JPH03201165A - Japanese language analytical processing system - Google Patents

Abstract

PURPOSE:To attain effective Japanese language analytical processing by referring the concept clasification of declinable word included in events or the semantic clasification or features of nouns in word order relation with the declinable word. CONSTITUTION:The morpheme analysis, syntax analysis and semantic analysis of a Japanese sentence transmitted to a translation procession part 20 through an input part 10 are executed while referring a file 30 stored with grammatical rules and a translation dictionary. An even relation deciding circuit 24 executes analytical processing in accordance with the concept sorts of respective events obtained from the circuit 21 while referring a connecting relation rule file 50 describing the connecting relation between plural events, a conversion processing means 25 converts respective words into English words based upon the obtained syntax information, semantic information and event relation information and then forming processing circuit 26 these English words are connected in accordance with an English grammar to form an English sentence and outputs the English sentence to an output part 40. Thus, more effective Japanese analytical processing can be attained.

Description

DETAILED DESCRIPTION OF THE INVENTION (a) Field of Industrial Application The present invention relates to a Japanese language analysis processing method in, for example, a machine translation system.

(b) Prior Art The concept of a general machine translation system is shown in FIG.

In the figure, l is an original human input device, into which the original text to be translated is input. Reference numeral 2 denotes a translation processing device, which performs translation processing of the original text by referring to the grammar rules 31 and translation dictionary 32 in the storage device i3. Reference numeral 4 denotes a translation output device, which displays or prints the translation.

In a conventional translation processing device using such a system, a sentence expressing a certain event that describes a person's action or the state of an object and a sentence expressing another event different from this one serve to connect the events. It was not possible to uniquely determine the connection relationship for input sentences that are connected by words such as conjunctions that form a .

In other words, when a certain event and another event are connected by a word that serves to connect the events, conventionally, the semantic connection between the events is often determined from only the first word; As a result, there was a fear that the translation would be inappropriate in subsequent translation processing.

Such concerns are particularly felt in Japanese language analysis when translating Japanese into foreign languages. Therefore, in order to compensate for this, it is necessary for the operator to edit the Japanese original text in advance in order to clarify the connections between events in the Japanese original text, which increases the burden on the operator. There was a drawback to that.

(c) Problems to be Solved by the Invention The present invention has been made in consideration of the above circumstances, and its purpose is to determine only the meaning of the words connecting the events when determining the connection relationship between events. The aim is to realize an effective analysis processing method for analyzing Japanese by clarifying the connection relationships between events from the semantic concept of the words of each event, rather than determining from the beginning.

To this end, we aim to realize a system that supports the construction of rules for determining connection relationships between events by registering rules for determining connection relationships between events, or by combining new rules from multiple rules. .

(d) Means for Solving the Problem The Japanese language analysis processing method of the present invention includes an input means for inputting a Japanese sentence, and an event expressed by the Japanese sentence inputted by the input means is included in the sentence. An event determination means that makes a judgment based on a predicate or a nominal that co-occurs with the predicate, and when a predicate is pronounced multiple times in an input Japanese sentence, the connection relationship between each event is acquired from the original text. and connection relationship registration means for registering connection relationships between acquired events.

Furthermore, the Japanese language analysis processing method of the present invention includes a human-powered means for inputting a Japanese sentence, and a predicate included in the text or a synonym for the phenomenon expressed by the Japanese text input by the input means. An event determination means for determining the connection relationship between the events expressed by each predicate; and a determination means for determining the connection relationship between the events expressed by each predicate; The connecting relation merging means is provided for merging subsequent relations.

(E) Effect According to the Japanese language analysis processing method of the present invention, when analyzing Japanese, information regarding the conceptual classification of the predicate included in the event and the semantic classification of the noun that co-occurs with the predicate is referred to. By doing this, it is possible to determine what contents are described by events that include predicates, and also to determine the connection relationships between events, thereby increasing the efficiency of Japanese language analysis. Furthermore, since it has means for registering new rules for determining connection relationships between events or means for merging new rules from existing connection relationship determination rules, it is easy to construct connection relationship determination rules. I can do it.

Therefore, in contrast to the conventional method that required complicated pre-editing processing by the operator in order to clarify the connections between events in a Japanese input sentence, the present invention uses the above processing method to clarify connections between events in the original text. It becomes possible to automatically and appropriately set relationships. Further, in the present invention, even when registering or merging rules for determining connection relationships between events, the processing efficiency is improved, so that the burden on the operator can be reduced.

(f) Embodiment of the Invention FIG. 1 shows a system configuration diagram for realizing the Japanese language analysis processing method of the present invention.

In the example below, “<Event l> + [Connect events!!]
The Japanese language analysis processing method of the present invention will be explained in detail by exemplifying the process of determining the connection relationship between two events of a Japanese sentence expressed in the form "+<event 2>".

In the system of the present invention shown in the figure, a Japanese sentence input from the input section 1O is transmitted to the translation processing section 20. In this processing unit 20, a basic analysis processing circuit 21 that performs morphological analysis, syntactic analysis, and semantic analysis performs each basic analysis process while referring to a file 30 that stores grammar rules and translation dictionaries. At this time, the expression contents of the input sentences ``Event 1'' and ``Event 2'' are determined by the conceptual classification of each term obtained from the dictionary as shown in FIG. 3 of the file 30.

Next, in order to determine the connection relationship between events with <event 1> and event 2>, the event relationship determination circuit 24 determines the connection between events according to the conceptual classification of each event obtained from the basic analysis processing circuit 21. Reference is made to the rules in the connection relationship rule file 50, which includes rules such as those shown in FIGS. 4 and 5 that describe the relationships.

That is, in the file 50, typical connection relationships between events are stored in advance as rules.
The inter-event connection relation that matches the combination condition with 〉special event 2〉 is the connection relation of this input sentence, and the file 5
Read from 0.

Based on the syntactic information, semantic information, and event relationship information obtained through each analysis process up to the event relationship determination circuit 24,
The conversion processing circuit 25 converts each word of the Japanese sentence into an English word, and the generation processing circuit 26 connects these English words according to English grammar to generate translated English sentences, and this generated English sentence is sent to the output section 40. displayed or printed out.

The above processing process is performed in steps Sl and S in the flowchart of FIG.
2, S3, S4, S5.5IO1Sll and 512.

On the other hand, if the inter-event connection relationship that matches the combination condition of the input text with event 1 and event 2 is not stored or registered in the connection relationship rule file 50 described above, it is necessary to register this relationship. If so, the operator can register this connection relation rule as necessary.

The first feature of the method of the present invention is that it is provided with a means for registering connection relation rules as necessary, and in this system, the event amount relation registration processing section 22 executes this.

For example, ``He earned money by working at a bookstore.''
The sentence, i.e. Event 1... ``Nami worked at the bookstore.''<Event2>
... When a document is input in which two events consisting of "He earned an income" are connected by the conjunctive particle "te (de)", the operation is described below based on the flowchart in Figure 2. Details are given below.

First, in step Sl, the above Japanese sentence is input to the input field lO.
Furthermore, in the basic analysis processing circuit 21, the grammar rules and translation dictionary file 30 are input in step S2.
Semantic analysis processing is performed with reference to . Subsequently, in step S3, the basic analysis processing circuit 21 determines whether there is a basic analysis connective word. In this case, by detecting the conjunctive particle "te" according to the grammar rules in file 30, the conjunctive particle M! Determine the presence or absence of I.

Then, in the basic analysis processing circuit 21, based on the dictionary in the file 30 as shown in FIG.
``Work'', its conceptual classification is [labor], the predicate of Event 2> is the verb "obtain", its conceptual classification is [material acquisition], and the noun that co-occurs with the predicate "obtain". It is detected that "income" is "income" and its meaning classification is "money".

Therefore, in step 4, the above-mentioned conjunctive particle ``te'', the verb ``work'' in the predicate of <Event 1>, the verb ``obtain'' in the predicate of <Event 2>, and the co-occurring noun Retains each “income” data. The retained data (referred to as RELI) in this case is as follows.

The conceptual relationship between events in this case (referred to as REL1-1) is expressed in the grammar rule translation dictionary file 3 as shown in Figure 3.
Based on information extracted in advance from 0, it is expressed as follows.

In this state, in step S5, the connection relation rule file 50 contains a rule that satisfies the above RELI condition, that is, the above R
If a connection relationship corresponding to the conceptual relationship of ELI-1 does not exist, the process advances to step S6 and waits for an instruction from the operator as to whether to newly register the connection relationship at this time. Note that, in step 5, if there is a rule that satisfies the above RELI condition, the process proceeds to steps 5tO1Sll and 512, as in the normal processing described above, and the translation processing is continued in sequence.

If there is a registration instruction from the operator in step S6, the process proceeds to step S7, where the event amount relationship registration processing section 22, which is a feature of the present invention, operates and registration processing is executed.

At this time, on the display screen W of the output unit 20, the above RELI
, and RELI-1 data, as shown in FIG.
Indication D2 of the conceptual classification [labor] of "work" and the conceptual classification [material acquisition] of "obtain" in <event 2> and its co-occurring noun "
A display D3 of the meaning classification [money] of "income" and a display D4 of various connection relationships between events are shown.

According to this display screen W, the operator selects various connection relationships appropriate for the input text from the display D4 by specifying the cursor or by specifying numbers; in this case, [2. By selecting [Cause/Reason], the memory map shown in Figure 4 will be displayed.
The RELI-1 information and the corresponding connection relationship information are registered in the connection relationship rule file 50. Specifically, as shown in Figure 4, a concept pointer is created in which the start address of the data area whose verb concept is [labor] and the data area whose verb concept is [material acquisition] are connected by [cause/reason]. The address is specified by .

In this way, when a connection relationship between events does not exist in the connection relationship rule file 50 that describes it, this connection relationship can be automatically newly established in the connection relationship rule file 50 according to an operator's instruction.

According to the connection relation rule file 50 in which the connection relation rules have been constructed as described above, from now on, the input text having the conceptual relation of RELI-1 will be processed in step SIO according to the memory map shown in FIG. The event relationship determination circuit 24
can determine that the connection relationship at this time is [cause/reason].

On the other hand, in step S5, the connection relation rule file 50
If there is no rule that satisfies the conditions of the input document, that is, a connection relationship corresponding to the conceptual relationship, and the operator does not specify registration in step S6, the system automatically creates a new rule from the existing rule in step S8. An attempt will be made to integrate the rules. If fusion is possible, secondary rules as described below can be obtained as a result of the fusion in step S9.

A second feature of the method of the present invention is that it includes a means for automatically merging and creating connection relation rules, and in this system, the event relation merging processing section 23 executes this.

Here, for example, the sentence ``He worked at a bookstore and was able to buy her a bouquet of flowers.'' In other words, <event l>...
・“He worked at a bookstore.”〈Event 2〉…A document in which two events consisting of “He was able to buy her a bouquet of flowers.” are connected by the conjunctive particle “te” (de). The operation will be described in detail below based on the flowchart of FIG.

The conditions between events in this case (referred to as REL2) are as follows, and based on this, the connection relation rule file 50
Conceptual relationships between events from the rules of (referred to as REL2-1)
As a result, a connection relationship corresponding to REL2-1 is automatically derived.

The conceptual relationship REL2-1 between events in this case (not registered in the connection relationship rule file 50) is based on information extracted in advance from the grammar rule translation dictionary file 30 as shown in FIG. It is expressed as below.

Then, the connection relation rule file 50 is searched for a conceptual relation having the same information on the first event as this, and as a result, a large number of conceptual relations including, for example, the above-mentioned RELI-1 are extracted.

Then, for each of the extracted conceptual relationships, a conceptual relationship whose first event has the same information as the second event is searched. For example, in the case of REL1-1 described above, a conceptual relationship having [substance acquisition] & [money] as the first event relationship is extracted.

Among the conceptual relationships obtained as a result, the information of the second event is the information of the second event of REL2-1 [Purchase] and the modal (
possible), and search for the same conceptual relationship as the conceptual relationship (possible), as shown below.
REL3-1) can be found, fusion is performed in step S9.

In step S8, the event relationship fusion processing unit 23
Based on the information [labor] for the first event in 2-1, this conceptual rule itself states that, for example, in response to the following event-same conditions (referred to as REL3), these events are [cause/reason]. This is the one that is originally registered as being connected.

In this case, the rules of REL3-1 are registered as shown in the memory map in Figure 5, and the connection relationship [cause/reason]
is detected.

That is, in step S9, [material acquisition] & [money]
Two conceptual relations RELI-1 and REL3-1 that are common
REL2-1 is fused from , and as a result, step S
In IO, as the connection relationship of REL2-1, REL
The connection relationship 3-1 [Cause/Reason] will be adopted.

Thereafter, normal translation processing continues at steps 811 and 512.

Further, in step S8, if merging is not possible, the process proceeds directly to steps 811 and 512 without determining the connection relationship in step Sll, and translation processing is performed with the connection relationship unknown.

(G) Effects of the Invention When the connection relationship between events is not clear, the Japanese language analysis processing method of the present invention can perform conceptual classification of predicates included in the event, semantic classification of nouns co-occurring with predicates, and classification of each noun. By referring to the information regarding the characteristics of the event, it becomes possible to determine the connection relationship between events, and by clarifying the connection relationship of events, effective Japanese language analysis processing becomes possible. Moreover, it is possible to register new judgment rules for connection relationships between events, or to create new rules by merging existing rules, which greatly reduces the burden on the operator, such as editing the original text. Since the groove bottom of the connection relationship determination rule in the connection relationship rule file can be simplified, system last can be significantly reduced.

[Brief explanation of drawings]

Fig. 1 is a ma diagram of an embodiment of a system that realizes the Japanese language analysis processing method of the present invention, Fig. 2 is a flowchart showing the processing operation of the method of the present invention, Figs. 3, 4, and 5. is a memory diagram, FIG. 6 is a schematic diagram of a display screen adopted in the method of the present invention,
FIG. 7 is a conceptual diagram of a general machine translation system. lO...Input unit, 20...Translation processing unit, 21...
Basic analysis processing circuit, 22...event relationship registration processing section, 2
3... Event relationship fusion processing unit, 24... Event relationship determination circuit, z5... Conversion processing device, z6... Generation processing device, 30... Grammar rule translation dictionary file, 40.
... Output section, 50... Connection relationship rule file.

Claims (3)

[Claims] (1) An input means for inputting a Japanese sentence, and an event in which the event expressed by the Japanese sentence inputted by the input means is judged based on the predicate included in the sentence or the noun that co-occurs with the predicate. A determination means, a support function for acquiring the connection relationship between each event from the original text when multiple predicates are included in the input Japanese sentence, and a connection relationship for registering the connection relationship between the acquired events. A Japanese language analysis processing method that includes a registration means. (2) An input means for inputting a Japanese sentence, and an event in which the event expressed by the Japanese sentence inputted by the input means is judged based on the predicate included in the sentence or the noun that co-occurs with the predicate. a determining means, a determining means for determining a connecting relation between events expressed by each term, and a connecting relation merging means for newly merging a connecting relation from the connecting relations between a plurality of events determined by the determining means. A Japanese language analysis processing method that is equipped with the following. (3) An input means for inputting a Japanese sentence, and an event in which the event expressed by the Japanese sentence inputted by the input means is judged based on a predicate included in the text or a nominal expression that co-occurs with the predicate. A determination means, a support function for acquiring the connection relationship between each event from the original text when multiple predicates are included in the input Japanese sentence, and a connection relationship for registering the connection relationship between the acquired events. a registration means, a determination means for determining the connectivity between events expressed based on each predicate or a nominal that co-occurs with the predicate; and a determination means for determining the connectivity between the plurality of events determined by the determination means and a connective relationship fusion means for fusing connective relationships into the input means, based on the event connective relationship registered by the connective relationship registering means or the connective relationship fused by the connective relationship fusion means. A Japanese language analysis processing method characterized by analyzing the connection relationships between multiple events in a sentence.