JPH08202715A - Machine translation system - Google Patents

Abstract

PURPOSE: To prevent ambiguity from being increased by the translation of a summerized sentence by excluding a node from a syntax analystic tree prepared by an analystic part based on fixed conditions. CONSTITUTION: A syntax analystic tree input part 1 receives the syntax analystic tree, for which the components of an input sentence are allocated to respective nodes, from the analystic part of the machine translating system and sends it to a node excluding part 2. The node excluding part 2 prepares a summerized tree by scanning the nodes but inquires it to an ambiguity deciding part 4 whether it is satisfied as conditions or not that the translation generated from the summerized tree contains just the same ambiguity as the ambiguity contained in the translation generated from the syntax analystic tree before summerization and the ambiguity deciding part 4 decides whether excluding the node or not by using knowledge stored in an ambiguity deciding knowledge storage part 5. After all the nodes are judged, concerning the remaining syntax analystic trees, a translating part translates the tree in a second language and the sentence in the second language is generated and outputted by a generating part.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation device for translating an input summary of a sentence in a first language into a sentence in a second language and outputting the sentence.

[0002]

2. Description of the Related Art A method for creating a summary sentence by sequentially deleting modifiers from an input sentence has been known.

In the technique disclosed in Japanese Patent Application Laid-Open No. 6-35961, an input sentence is input to the abstracting device in the form of a parse tree as shown in FIG. Each component of the input sentence is assigned to each node of the syntactic analysis tree. Further, a level representing the depth of the tree is assigned from the root node to the leaf node. The root node (level 1) is
It is a component "providing" that controls the entire input sentence.
To the child node (level 2) of the root node, the three components "system", "function", and "us" that are controlled by the root node are assigned. Leaf nodes (level 3) are "this", which is dominated by "system is", "two" and "convenient", which are dominated by "function".

The abstraction is realized by extracting the components assigned to nodes at a level lower than a certain level. In other words, if only the nodes of level 1 are extracted, the summary sentence becomes “provided”, and if the nodes of a level shallower than level 2 are extracted, “the system provides us with the function”. .

[0005]

However, when the conventional method is directly incorporated in the machine translation device, the following problems occur. In other words, the ambiguity that does not occur if you translate the input sentence as it is without summarizing,
This may occur when translating the summary of the input sentence. This problem is not mandatory in the first language (it can be omitted),
It occurs when information that is mandatory (cannot be omitted) in the second language is deleted. This is because the conventional summarizing device has linguistic knowledge about the first language, but does not have linguistic knowledge about the first language and the second language.

As an example, let us consider a case where the first language is Japanese and the second language is English, and "This system provides us with two convenient functions" is input. Not required in the first language (Japanese) above, but in the second language (English)
In this case, the information that is indispensable includes the clarification of each element of the verb, the distinction of the number of noun phrases, and the distinction / indefinite distinction of noun phrases.

If the input sentence is translated as it is without being summarized, there are two ways as shown in FIG. In contrast,
When the summary sentence “provide.” Obtained by extracting the level 1 nodes from the syntactic analysis tree of FIG. 4 is translated, there are 49 ways as shown in FIG. Further, when the summary sentence “The system provides us with a function” obtained by extracting nodes at a level shallower than the level 2, the number is increased to 16 as shown in FIG.

Generally, in Japanese, even a single verb (for example, "provide") is valid as a sentence, but it is not valid in English. In order not to violate English grammar, case elements such as subject and object are necessary. In other words, in order to translate a sentence "provide." That does not violate Japanese grammar so that it does not violate English grammar, the subject and object of "supply", which is a translation of the verb "provide", should be changed to a pronoun, etc. Must be specified by. The reason why the possibility of translating the abstract "provide." Is increased is that the subject "system" and the object "function", which were specified in the sentence before the abstract, are deleted.

Usually, one Japanese noun phrase (for example,
When translating "convenient functions" into English noun phrases, singular indefinite noun phrases ("a convenient function"), singular definite noun phrases ("the convenient function"), plural indefinite noun phrases ("convenient functions") There are four possibilities: ") and plural definite noun phrases (" the convenient functions "). The reason for increasing the possibility of translating the abstract sentence "The system provides us with function" is the noun phrase "this system".
The demonstrative "kono" and the noun phrase "two useful functions" that provided information for distinguishing the number of each and the definite / indefinite.
This is due to the deletion of the adnominal “two” which provided information to distinguish the number of.

[0010]

In order to solve the above problems, the present invention provides an input unit for receiving a sentence input in a first language, a morphological analysis and a syntax for an input sentence input from the input unit. An analysis unit that performs parsing and semantic analysis to create a parse tree, and a translation generated from the parse tree from the parse tree created by the parse unit is included in the translation generated from the parse tree before summarization. A summarization unit that creates a summarization tree by deleting a node on condition that it contains only the same ambiguity as the ambiguity, and a knowledge storage for ambiguity determination that stores knowledge for determining whether to delete the node in the summarization unit. Section, a converting section for converting the summary tree summarized in the summarizing section into a tree in the second language, a generating section for generating a sentence in the second language from the converting section, and an output for outputting the sentence generated by the generating section. Equipped with a.

[0011]

In the machine translation device with the summarization function according to the present invention, in the normal machine translation device, the summarization unit is provided after the processing of the analysis unit in the process of input unit → analysis unit → conversion unit → generation unit → output unit. There is. The summarizing section includes summarizing executing means and ambiguity determining means.

The summary executing means determines whether or not to delete the node according to the procedure of the forward scan on the parse tree,
Create a summary tree. As shown in FIG. 10, the forward scan first scans the root node of the tree T, and then scans the subtree T1 having n child nodes of the root node of the tree T as root nodes. That is, T2, ..., Tn are scanned in this order. The summary executing means inquires of the ambiguity determining means as to whether or not the condition shown in FIG. 11 is satisfied, and the answer to the ambiguity determining means determines whether or not to delete the node.

The ambiguity determination means compares with the ambiguity determination knowledge storage unit whether the summary tree satisfies the condition of FIG. Answer. Here, when each condition as shown in FIG. 12 is met, the node is not deleted because it is essential information for its parent node.

In this way, after making judgments on all the nodes, the conversion unit converts the remaining parse tree into the second language tree, and the generation unit converts the sentence in the second language. Is generated. The generated summary sentence is output from the output unit.

[0015]

FIG. 2 shows the functional configuration of a summarizing section (summarizing device) of a machine translation apparatus according to the present invention. In the summarizing device, the parse tree input unit 1 receives a parse tree in which constituent elements of an input sentence are assigned to each node from the analysis unit of the machine translation device and sends it to the node deleting unit 2.

The node deleting unit 2 scans the nodes of the parse tree according to the procedure of the forward scan to create a summary tree. (Processing of the nodes when the parse tree of FIG. 4 is scanned in the forward order) The order is shown in FIG. 10), and the ambiguity determination unit 5 is inquired whether or not the summary tree satisfies the condition of FIG. 11, and whether or not to delete the node is determined according to the answer.

The summary tree output unit 3 sends the summary tree received from the node deletion unit 2 to the conversion unit of the machine translation device. The ambiguity determination unit 4 uses the knowledge stored in the ambiguity determination knowledge storage unit 5 to determine whether or not the summary tree received from the node deletion unit 2 satisfies the condition of FIG. 11, and the determination result Is notified to the node deletion unit 2. The ambiguity determination knowledge storage unit 5 stores knowledge (FIG. 12) for checking whether or not the condition of FIG. 11 is satisfied.

The flow of processing in the summarizing device according to the present invention is shown in FIG. The flow of processing will be described below by taking the case of creating a summary tree from the syntax analysis tree of FIG. 4 as an example.

At STEP 1, the syntax analysis tree is received from the analysis unit of the machine translation device, and at STEP 2, the root node is processed.

In STEP 3, it is checked whether the scanning of the nodes can be finished. Since the root node is not targeted for deletion, the processing ends if the parse tree consists of only root nodes. Now, the root node “provide” has child nodes, and since they have not been scanned, go to STEP4.

In STEP 4, the node to be processed is moved from the current node to the next node according to the procedure of the forward scan. According to the procedure of forward scan, the nodes to be processed are “provided”, “system is”, “this”, “function”,
Go to "two", "convenient", "to us", in that order.

Next, in STEP 5, the current processing target node N
And the parent node M are compared with the knowledge in FIG.
In TEP6, it is determined whether the condition of FIG. 11 is satisfied even if the node N is deleted according to the collation result. Here, the condition of FIG. 11 being satisfied means that the node N and the parent node M are
Is not consistent with the knowledge of FIG. If the condition is satisfied, in STEP 7, the subtree having the node N as a root node is deleted. In FIG. 12, SBJ
Means a subject, OBJ means an object, and NP means a noun phrase. An English translation of a Japanese phrase (for example, the English translation "supply" of "provide") is obtained by a bilingual dictionary.

Now, the node N to be processed is "system" and its parent node M is "provide." The set of nodes “system” and “provide.” Is shown in FIG.
The node "system is" cannot be deleted because it matches the second line of knowledge of No. 2 (the node is an essential case of the parent element).

Returning from STEP 3 to STEP 4, "this" is set as the processing target node N. The parent node M of N is “system is”. The set of nodes "this" and "system is"
The node “this” cannot be deleted because it matches the third line of knowledge in FIG. 12 (the node has a role of fixing the parent node to the singular / zero article).

Furthermore, returning from STEP3 to STEP4, "function" is set as the processing target node N. The set of the node “function” and its parent node “provide.” Matches the first line of knowledge in FIG. 12 (the node is an essential case of the parent element).
Therefore, the node "function" cannot be deleted.

Thereafter, similarly, a set of nodes “two” and its parent node “function”, a set of nodes “convenient” and its parent node “function”, a node “to us” and its parent node “provide” To do. "
12 are sequentially compared with the knowledge in FIG. Node "two"
And the parent node "functional" set coincides with the fourth line of knowledge in FIG. 12, the node "two" cannot be deleted. When the collation of the set of the node "convenient" and its parent node "function" and the knowledge in FIG. 12 is attempted, none of the lines match. Therefore, even if the subtree whose root node is "convenient" is deleted, the condition of FIG. 11 is satisfied, so this subtree is deleted. Further, since the set of the node “to us” and its parent node “to provide” does not match any row of the knowledge in FIG. 12, the subtree having the node “to us” as the root node is deleted.

After all the nodes have been scanned, the summary tree shown in FIG. 9 is finally obtained, and the translation thereof is as shown in FIG. 8, which has the ambiguity of the translation of the sentence before the summary (FIG. 5). They will only have the same ambiguity.

[0028]

As described above, according to the present invention,
The input sentence can be summarized such that the translation of the sentence that summarizes the input sentence contains only the same ambiguity that is included in the translation of the input sentence before the summary. That is, it is possible to prevent the ambiguity of the verb case element and the ambiguity of the noun phrase from increasing by translating the summary sentence.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of a machine translation device according to an embodiment of the present invention.

FIG. 2 is a schematic block diagram of a summarizing device according to an embodiment of the present invention.

FIG. 3 is a diagram showing a flow of a digest process in an embodiment of the present invention.

FIG. 4 is a diagram showing an example of a parse tree of a sentence “This system provides us with two convenient functions.”

FIG. 5 is a diagram showing possible translation examples for the sentence “This system provides us with two convenient functions.”

FIG. 6 is a diagram showing a possible translation example of a summary sentence “to be provided” according to a conventional technique.

FIG. 7 shows an example of possible translations for the abstract “The system offers us functionality” according to the prior art.

FIG. 8 is a diagram showing possible translations of a summary sentence “This system provides two functions.” In an embodiment of the present invention.

9 is a diagram showing a summary tree obtained by the present invention from the parse tree shown in FIG. 4;

FIG. 10 is a diagram showing a scanning order of a syntax analysis tree.

FIG. 11 is a diagram showing a condition to be satisfied by a summary tree in the present invention.

FIG. 12 is a diagram showing an example of nodes to be included in a summary tree according to the present invention.

[Explanation of symbols]

 1 parsing tree input unit 2 node deletion unit 3 summary tree output unit 4 ambiguity determination unit 5 ambiguity determination knowledge storage unit

Claims (5)

[Claims] 1. An input unit that receives a sentence input in a first language; an analysis unit that performs morphological analysis, syntactic analysis, and semantic analysis on the input sentence received from the input unit to create a syntactic analysis tree; From the syntactic analysis tree created by the parsing unit, a summarization unit for creating a summarization tree by deleting nodes based on certain conditions, and converting the summarization tree summarized in the summarization unit into a second language tree. A machine translation apparatus comprising: a conversion unit, a generation unit that generates a sentence in a second language from the conversion unit, and an output unit that outputs the sentence generated by the generation unit. 2. The summarization unit stores knowledge, provided that the translation generated from the summary tree includes only the same ambiguity as the ambiguity included in the translation generated from the pre-summarization parse tree. An ambiguity determination knowledge storage unit is provided, and each contact point is collated with the ambiguity determination knowledge storage unit.
The machine translation apparatus according to claim 1, further comprising a node deletion unit that determines whether to delete the node based on the answer. 3. The ambiguity determination knowledge storage unit,
3. The machine translation device according to claim 2, wherein knowledge for determining whether a node is an indispensable case element of the parent contact's dialect is stored, and the node is not deleted when the node corresponds. 4. The ambiguity determination knowledge storage unit,
3. The machine translation device according to claim 2, wherein knowledge for determining whether or not a node is an element that is the number information of the noun phrase of its parent contact is stored, and if the node corresponds, the node is not deleted. . 5. The ambiguity determination knowledge storage unit,
The knowledge for determining whether or not a node is an element that serves as information for distinguishing between fixed and indefinite of a noun phrase of its parent contact is stored, and the node is not deleted when the node corresponds. The machine translation device described.