JP2723886B2 - Machine translation apparatus and translation rule creation method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a translation rule creation method using translation processing in a machine translation method using translation rules registered in a storage device in advance. 2. Description of the Related Art In natural language machine translation, it is common to register translation rules such as grammar rules of a target language in a storage device in advance and perform translation while referring to the rules. The translation rule basically defines a translation based on the form of the target sentence. [Problems to be solved by the invention] It is desirable to prepare all translation rules before performing the translation process. However, natural languages have many exceptional grammatical rules and special shampoos. It is a common practice to augment the rules while translating experimentally.
In addition, there are many sentence patterns unique to the target field, and translation rules may need to be added according to the needs of the user of the translation system. These translation rules are described, for example, in Okamoto et al., "Study of Method of Syntax Ideom in English-Japanese Machine Translation System ATHENE / E," Proc. To create translation rules, recognize missing rules,
A large amount of manual work is required, such as writing and registering rules and checking for correct rules. In particular, it is difficult to write a rule that has conditions that correctly match a sentence that requires the rule. For example, Nakamura et al., "Grammar Development Tools for Machine Translation Using Multi-windows"
As described in the IPSJ 29th National Convention Preliminary Proposal, these tools can be used directly to collect and search information for extracting problems in existing rules and to edit data for describing rules. Does not create rules. An object of the present invention is to provide a method for easily creating a translation rule that correctly operates on a sentence requiring a translation rule by using a machine translation process in view of the above points. [Means for Solving the Problems] In order to achieve the above object, the present invention provides a method of translating a target sentence, displaying an original sentence and a translation of a translation result on a screen,
If a new translation rule is required, use the manual intervention operation and the translation process to convert the original and translated sentences to create a translation rule, re-translate using that rule, and check the result on the screen After that, the translation rule is registered. [Action] In the present invention, the translation rule is re-translated using the translation rule, the result is confirmed on the screen, and then the translation rule is registered. Can be created. EXAMPLES Hereinafter, the present invention will be described in detail with reference to examples. As an example, a method of creating a translation rule used when performing a translation process from English to Japanese will be described. In FIG. 2, 1 is a processor, 2 is a memory for English text to be translated, 3 is a dictionary memory used for translation, 4 is a memory for translation rules, 5 is an internal table, 6 is a display, and 7 is a keyboard. Show. The outline of creating a translation rule will be described with reference to a flowchart shown in FIG. First, the processor 1 has an English text memory 2
Reads the English sentence to be translated into the internal table 5 from
A dictionary for translation registered in the dictionary memory 3 in advance;
The relevant part of the translation rules registered in advance in the translation rule memory 4 is read into the internal table 5, and the translation is performed by referring to it (step 101). Processor 1
Information such as the analysis result of the English sentence created and used at the time of translation is saved in the internal table 5, and the translated English sentence and a translated sentence of the result are displayed on the display 6 (step 102). The creator of the translation rule looks at the translation on the display 6 and determines whether or not a new translation rule is necessary (step 103).
When a new translation rule is required, the processor 1 uses the translation rule creator's keyboard 7 operation, the information at the time of translation in the internal table 5 and the translation process to display the translation rule on the display 6. Is created (step 104). The newly created translation rule is registered in the translation rule memory 4 (step 105). When all the translation rules have been created, it is checked whether or not the English sentence to be translated still remains in the English text memory 2 (step 106). If there is, the process returns to step 101. Next, steps 103 to 105 for creating a translation rule will be described in detail. As an example,
A description will be given of a case where a translation rule (hereinafter, referred to as a syntax pattern rule) in a format represented by a pair of an English sentence pattern and a translated sentence pattern is created. (1) Syntax Pattern Rule The syntax pattern rule is stored in the translation rule memory 4 in the form of, for example, a syntax pattern rule table as shown in FIG. The syntax pattern rule table includes an English sentence pattern area 31 for storing an English sentence pattern to be treated as a pattern, a translated sentence pattern area 32 for storing a pattern translated from the pattern, and a pattern information area 33 for storing information on the entire pattern. Become. In the English sentence pattern, an English word and a keyword (N, VP, SENT, etc., hereinafter, referred to as a grammatical symbol) representing a grammatical meaning of a group of words, such as a part of speech, a verb phrase, a sentence, are described side by side. The conditions of various attributes (grammatical attributes, semantic attributes, etc.) for the grammatical symbol are described in parentheses after the grammatical symbol. In the translated sentence pattern, the translated word itself and the corresponding grammatical symbol in the English sentence pattern are described side by side. Conditions for translating grammatical symbols, such as verb conjugation endings, are described in parentheses after the grammatical symbols. If the same grammatical symbol is found in one pattern, it is distinguished by numbering according to the order in which it appears in the English sentence pattern, such as NP1 and NP2.
As the pattern information, a grammatical symbol is described and its attribute is enclosed in parentheses in order to specify how a group of patterns is treated in an English sentence. For example, the first line in FIG. 3 contains the English word “It” and the English word “t”.
akes ", a noun phrase representing a person, a noun phrase representing a time, the English word" to ", and a verb phrase appear in this order in the English sentence,
It represents a translation rule that recognizes that part as a syntax pattern and translates it according to the translation pattern. For example, the entire English sentence “It takes me an hour to go to school.” Is recognized as a syntactic pattern, the entire pattern is interpreted as a positive sentence according to the pattern information, and “I go to school” according to the translated sentence pattern. It takes an hour. " As shown in the third row of FIG. 3, the entire syntax pattern may be a noun phrase instead of a sentence. In this way, a good rule that cannot be handled by general translation rules is described one by one as a syntax pattern rule. (2) Analysis table The example sentence "The wisest man cannot know everything." Is translated into the sentence "The wisest person cannot know everything." At this time, for example, an analysis result of an English sentence as shown in FIG. 4 is created and used for translation, and is stored in the form of an analysis table as shown in FIG. 5, for example.
Each word is numbered to represent the analysis result. 4th
In FIG. 5 and FIG. 5, this is indicated by W1 to W6. Examples of words that are continuous and grammatically meaningful in English sentences are:
Collect them as a single node, and number them.
In FIGS. 4 and 5, this is indicated by N1 to N9. Each word forms one node. The analysis table in FIG. 5 stores information for each node, and includes a node number area 51 for storing a node number,
Of the words that make up the node, the leftmost word number area 52 that stores the word number of the leftmost word in the English sentence, the rightmost word number area 53 that also stores the word number of the rightmost word, the grammatical symbol of that node Grammar symbol area 5 indicating
4. It has an attribute area 55 for storing the attribute of the node.
In addition, the translation start position area 56, which is stored at the time of translation generation, in which character in the translation the translation of the node starts from,
It also has a translated word end position area 57 for storing the number of the end of the character. For example, the node N1 is composed of only the word W1 and is a definite article indicating that there is no corresponding translation. The node N7 is a noun phrase composed of the words W1 to W3 and representing a single person, and indicates that the translated word is "the smartest person". (3) Judgment of Whether to Create Rules According to the flowchart of FIG. 1, the analysis table of FIG. 5 is stored in the internal table 5 of FIG. Is displayed on the display 6. The display in FIG. 6 includes a translation result display section 61 for displaying a translated English sentence and its translation, a rule display section 62 for displaying a translation rule being created,
It consists of three parts, a message display section 63 for displaying guidance to the translation rule creator and displaying error messages. In FIG. 6, nothing is displayed on the rule display section 62 yet. Guidance on which rule to create is displayed on the message display section 63. FIG. 7 shows an example of the keyboard 7 used in the present invention. The keyboard has alphabet / kana keys for entering alphabetic characters, kana characters, and kanji, and symbol keys for entering symbols, special keys for switching kanji / kana keys, inserting and deleting characters, and a message display section for display. There are function keys (F1 to F7) for selecting a process according to 63. In FIG. 6, when the function key F7 in FIG. 7 is pressed, the processor 1 determines that a new translation rule is not necessary, and proceeds to the determination as to whether or not to translate the next sentence.
If any other function key is pressed, the process proceeds to the process of creating a translation rule for the corresponding number in the message display section. (4) Creation of Rule In FIG. 6, when the function key F3 is pressed, the processor 1 displays, for example, a screen as shown in FIG. FIG. 8 shows that an English sentence pattern slot 621, a translated sentence pattern slot 622, and a pattern information slot 623 are displayed on the rule display section 62. The English sentence pattern slot 621 and the translated sentence pattern slot 622 respectively display the same English sentence and translated sentence as the translation result display section 61. Nothing is displayed yet in the pattern information slot 623. The message display section 63 displays the name of a processing mechanism used for creating a syntax pattern rule. As shown in FIG. 9, in the English pattern slot 621,
When "<" and ">" are inserted by the insertion key and the symbol key of the keyboard 7 and the F1 key is pressed, the processor 1 determines that the target range of the syntax pattern rule is specified. Then, only the portion surrounded by “<” and “>” is displayed in the English pattern slot 621. Then the fifth
Referring to the leftmost word number area 52 and the rightmost word number area 53 of the analysis table shown in the figure, a node corresponding to the range displayed in the English pattern slot 621 is searched, and the translation start position area 55 and the translated word end position area 56 are searched. By referencing, only the translation corresponding to the node is displayed in the translation pattern slot 622, and referring to the grammar symbol area 54 and the attribute area 55, the grammatical symbol and the attribute possessed by the node are replaced with the grammatical symbol (attribute ...) And displayed in the pattern information slot 623. In the case of FIG. 9, since the range from the first word to the sixth word is set,
It corresponds to the node N9 on the analysis table in FIG. 5 and is displayed as a rule display section 62 in FIG. If there is no node corresponding to the analysis table, a message to that effect is displayed on the message display unit 63, and no other processing is performed. As a method of specifying the range in the English sentence, instead of "<"">",
The range may be underlined. As shown in FIG. 11, when "<" and ">" are inserted into the English pattern slot 621 and the F2 key is pressed, the processor 1 specifies a portion to be described with a grammatical symbol in the English pattern. Judge that it was done. Then, the English pattern slot 621 is examined in order, a node corresponding to the portion enclosed by “<” and “>” is searched from the analysis table, and the English pattern slot 621 and the portion are grammatical symbols (attributes ...). Replace with the form Referring to the analysis table, the corresponding part of the translated sentence pattern slot 622 is also replaced with the same grammatical symbol. In the case of FIG. 11, the nodes N2 and N8 are replaced with grammatical symbols and displayed as shown in FIG. If “<” and “>” do not correspond to each other or the corresponding node does not exist, an error message to that effect is displayed. After the conversion, as shown in FIG. 13, the translation rule creator uses the keyboard 7 to appropriately edit each slot in the rule display section 62, such as correcting the translation, deleting extra attributes, and inserting translation conditions. To change. When the F3 key is pressed, the processor 1 displays the rule display section 62
Is stored in the internal table 5 once,
Re-translation is performed on the assumption that the syntax pattern rule is already registered as a translation rule, and the result is displayed on the translation result display section 61 as shown in FIG. (5) Registration of rules When the F4 key is pressed, the processor 1 displays the rule display section 62.
Is registered in the syntax pattern rule table in the translation default memory 4. At this time, the English sentence pattern slot 621 corresponds to the English sentence pattern area 31, the translated sentence pattern slot 622 corresponds to the translated sentence pattern area 32, and the pattern information slot 623 corresponds to the pattern information area 33. When the F5 key is pressed, the processor 1 determines that the syntax pattern rule creation processing has been completed, and proceeds to a determination as to whether or not to perform the next translation. [Effects of the Invention] As described above, according to the present invention, since a translation rule is created while analyzing a sentence that requires a new translation rule, a rule that surely conforms to the sentence can be easily created. it can. In addition, before a created rule is officially registered as a translation rule, a translation experiment can be performed using the rule, so that affirmation of creation and confirmation of a translation rule can be shortened.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a flowchart of a translation rule creation process according to the present invention;
FIG. 2 is a block diagram of an apparatus for performing the translation rule creation processing of the present invention, FIG. 3 is a diagram showing an example of a syntax pattern rule table, FIG. 4 is a diagram showing an example of an analysis result created at the time of translation, and FIG. FIG. 6 shows an example of an analysis table, FIG. 6 shows a display example of a translation result, FIG. 7 shows an example of a keyboard, FIG. 8 shows an example of a translation rule creation screen,
FIGS. 9, 10, 11, 12, and 13 show examples of screens at each stage when a translation rule is created, and FIG. 14 shows a translation result when re-translation is performed. It is a figure showing the example of a display of. 1 ... Processor, 2 ... Memory for English text, 3 ...
… Dictionary memory, 4… translation rule memory, 5… internal table, 6… display, 7… keyboard, 31
…… English pattern area, 32 …… Translation pattern area,
33 …… Pattern information area, 51 …… Node number area,
52: Leftmost word number area, 53: Rightmost word number area, 54: Grammar symbol area, 55: Attribute area, 56 ...
Translation start position area, 57 …… Translation end position area, 61…
... Translation result display section, 62 ... Rule display section, 63 ... Message display section, 621 ... English pattern slot, 622 ... Translation pattern slot, 623 ... Pattern information slot.

Claims (1)

(57) [Claims] First storage means for storing a translation rule in the form of a syntax pattern expressed as a pair of a syntax pattern of a first natural language sentence and a syntax pattern of a second natural language sentence; A second storage means for storing a sentence of a natural language, a dictionary for storing words used for translation, a translation rule in a syntax pattern format of the first storage means and a word of the dictionary with reference to the words of the dictionary. Second natural language sentence
Processing means for translating into a sentence of a natural language, display means for displaying a bilingual sentence of the first natural language sentence and a sentence of the second natural language sentence, and input means. A phrase is specified by the input means with respect to the sentence of the first natural language, and a phrase of the specified first natural language sentence and a second phrase corresponding to the phrase are received.
And the syntax pattern of the sentence of the first natural language and the syntax pattern of the sentence of the second natural language in which the phrase of the natural language is replaced by a symbol indicating a grammatical role. When the syntax pattern of the language sentence is changed by the input means, a pair of the syntax pattern of the first natural language sentence and the changed syntax pattern of the second natural language sentence is converted into a new syntax A machine translation apparatus characterized in that the translation rule is registered in the first storage unit as a translation rule in a pattern format. 2. 2. The method according to claim 1, wherein said first storage means stores a syntax pattern of a sentence in a first natural language and a second
A machine translation apparatus comprising a translation rule table in a syntax pattern format representing pairs of syntax patterns of natural language sentences. 3. The processing unit refers to the storage unit that stores the translation rule in the syntax pattern format represented by the pair of the syntax pattern of the first natural language sentence and the syntax pattern of the second natural language sentence. A translation rule creating method for a machine translator for translating a sentence in a language into a second natural language sentence, comprising: displaying a bilingual sentence of a first natural language sentence and a second natural language sentence on display means; When a phrase included in the displayed first natural language sentence is specified using the input means, the specified first natural language sentence phrase and the second natural language phrase corresponding to the specified phrase The syntax pattern of the first natural language sentence and the syntax pattern of the second natural language sentence in which are replaced by symbols are displayed on the display means, and the displayed syntax of the second natural language sentence is displayed. The pattern is changed by interactive operation on the display means A pair of the syntax pattern of the first natural language sentence and the changed syntax pattern of the second natural language sentence is generated as a translation rule in a syntax pattern format, and the generated translation rule in a syntax pattern format is generated. Is registered in the storage means. 4. 3. The translation rule table according to claim 3, wherein said storage means stores a syntax pattern of a sentence of a first natural language and a syntax pattern of a sentence of a second natural language as a pair. A translation rule creating method for a machine translation apparatus, characterized in that: 5. 4. The method according to claim 3, wherein when the processing unit generates a translation rule in a syntax pattern format, a phrase of a designated first natural language sentence and a second natural language corresponding to the phrase A translation rule in the form of a syntactic pattern by adding a grammatical symbol indicating an attribute to the phrase of the sentence (1). 6. 4. The method according to claim 3, wherein a word in the translated second natural language sentence is corrected using the input unit.