JP3253311B2 - Language processing apparatus and language processing method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a language processing apparatus and a language processing method used for a machine translator, a Japanese summarization machine, a natural language interface of a computer or a robot controller, and the like.

[0002]

2. Description of the Related Art Japanese characters include kanji and kana. The above-mentioned kanji is an ideographic character and represents meaning in addition to reading. On the other hand, the kana is a phonetic character and represents only reading. Therefore, when analyzing or understanding the input Japanese sentence, the meaning of the kana written portion cannot be specified as it is. This is because a word represented by a kanji character that is an ideographic character has a narrower meaning than a word represented by a kana character that is a phonetic character. That is, for example, the following three words, (A) raise (B) raise (C) raise, are all given as kana. Therefore, when a Japanese sentence saying "I gave tempura" was input, the meaning of the word "given" was (a) given above (b) given to a person (c) fried Can have at least three different meanings. For this reason, the meaning of the word "given" written in the pseudonym cannot be specified. Therefore, usually in such a case,
It specifies the meaning of the kana written word of the Japanese sentence that was input by the semantic analysis using the case pattern. The above-mentioned case pattern is the case that a certain verb can take (the nominative case with the case particle "ga", etc., the target case with the case particle "wo", etc., the target case with the case particle "ni", etc., ...). It is associated with a semantic feature related to the nominative that fills the case slot. For example, rated pattern of the verb "buy" is to buy a "person" is "things", rated pattern of the verb "give" is given to the "people" is "things" to "people".

[0003]

However, when the meaning of a kana-written word is specified by the method using the case pattern, there are the following problems. That is, for example, if the kana writing word is a verb "Call", from the part of speech and semantic features of each word of the input sentence, rank pattern of the kana writing the word "put" has "human" to "what""Let's say that it is decided to hang on “Place”. However, as verbs having the kana notation "kake" and the case pattern, for example, there are "kake" and "kake". for that reason,
It is not possible to identify whether the meaning of the kana-written word “Kake” is “Kake” or “Kake” using only the determined case pattern. Therefore, the ambiguity of the meaning remains in the part of the kana written word in the input sentence, and the accuracy of subsequent processing is reduced, and a processing result different from the one originally intended by the operator is output. It is. As a result, in the case of a machine translation device, a phenomenon such as outputting a translation with an unknown meaning appears,
Since a great deal of labor such as post-editing is required, there is a problem that efficient translation processing cannot be performed.

It is an object of the present invention to provide a language processing apparatus and a language processing method which enable efficient language processing by reducing the ambiguity of the meaning of kana-written words in Japanese. .

[0005]

According to a first aspect of the present invention, there is provided a language processing apparatus for inputting a Japanese sentence, and a display unit for displaying the input Japanese sentence and the like. The input Japanese sentence is divided into a plurality of Japanese morphemes, a morphological analysis is performed, a morphological analysis unit that recognizes the word and its part of speech, a reading of the word, a plurality of kanji notations having the reading, and In the case where a kana-written word is present in a word string divided by the morphological analyzer,
A notation information processing unit that searches for readings of the kana written word from the readings stored in the notation information storage unit and displays a plurality of kanji notations associated with the searched reading of the word on the display unit And a kanji notation determining unit for designating and confirming a desired kanji notation from among the plurality of kanji notations displayed by the notation information processing unit, and the kana determined by the kanji notation determining unit. It is characterized in that the meaning of the kana written word is specified based on the kanji notation of the written word.

According to a second aspect of the present invention, there is provided a language processing apparatus comprising: an input unit for inputting a Japanese sentence; a display unit for displaying the input Japanese sentence; The morpheme is divided into morphemes, morphological analysis is performed, and a morphological analysis unit that recognizes the word and its part of speech, the reading of the word, a plurality of kanji notations having the reading, and the meaning of each kanji notation are associated with each other. When there is a kana written word in the stored notation information storage unit and the word string divided by the morphological analysis unit, search for the kana written word reading from the readings stored in the notation information storage unit. Then, a notation information processing unit that displays a plurality of kanji notations associated with the retrieved reading of the word on the display unit, and a desired kanji notation displayed from the plurality of kanji notations displayed by the notation information processing unit. Specify Kanji notation and confirm Kanji notation determination unit for
Based on the semantics of the kanji notation determined by the kanji notation determination unit and the analysis result of the morphological analysis unit, a syntax analysis unit that performs a syntax analysis of the Japanese sentence, and based on the analysis result of the syntax analysis unit, It is characterized by having a translation unit for translating the input Japanese sentence into a foreign language sentence.

According to a third aspect of the present invention, in the language processing apparatus of the first or second aspect, a kana writing frequency is added to each of the kanji notations stored in the notation information storage unit. When displaying the plurality of kanji notations, the notation information processing unit preferentially displays the added kanji notation with a high kana writing frequency. .

According to a fourth aspect of the present invention, there is provided an input means for inputting a Japanese sentence, a display means for displaying the input Japanese sentence or the like, a reading of a Japanese word and a plurality of kanji notations having the reading. A language processing method by a Japanese processing device having processing means for processing the input Japanese sentence based on the registration information, and registering the meanings of the respective kanji notations with each other in a memory means. The divided Japanese sentence is divided into a plurality of Japanese morphemes, morphological analysis is performed to identify the words and their parts of speech, and it is determined whether or not there are any kana-written words in the divided word strings. If it is determined that there is a Japanese word, the kana writing word is searched from the registered Japanese word readings, and a plurality of kanji notations associated with the searched word reading are searched. Display and this displayed multiple Selected by indicating a desired Chinese characters from among the Chinese characters, on the basis of the designated Chinese characters of the kana writing words, it is characterized by specifying the meaning of the kana writing words.

According to a fifth aspect of the present invention, there is provided an input means for inputting a Japanese sentence, a display means for displaying the input Japanese sentence, etc., a reading of a Japanese word and a plurality of kanji notations having the reading. A language processing method by a Japanese processing device having processing means for processing the input Japanese sentence based on the registration information, and registering the meanings of the respective kanji notations with each other in a memory means. The divided Japanese sentence is divided into a plurality of Japanese morphemes, morphological analysis is performed to identify the words and their parts of speech, and it is determined whether or not there are any kana-written words in the divided word strings. If it is determined that there is a Japanese word, the kana writing word is searched from the registered Japanese word readings, and a plurality of kanji notations associated with the searched word reading are searched. Display and this displayed multiple Of the input Japanese sentence based on the semantics of the specified kanji notation and the analysis result of the morphological analysis unit. Characteristically, the input Japanese sentence is translated into another language sentence based on the analysis result.

According to a sixth aspect of the present invention, in the language processing method according to the fourth or fifth aspect, each of the registered kanji notations is added with a kana writing frequency. When displaying a plurality of kanji notations, the kanji notation with a high kana writing frequency attached is preferentially displayed.

[0011]

According to the first aspect of the present invention, the input Japanese sentence is divided into a plurality of Japanese morphemes by the morphological analysis unit, and the morphological analysis is performed to identify the words and their parts of speech. Then, when there is a kana-written word in the word string divided by the morphological analysis unit, the notation information processing unit searches for the reading of the kana-written word from the notation information storage unit. A plurality of kanji notations associated with the reading of the word are displayed on the display unit. Then, the kanji notation determination unit is operated by the operator, and from among the plurality of kanji notations displayed on the display unit,
The desired kanji notation is instructed and confirmed. Thus, the kanji notation of the kana-written word in the input Japanese sentence is determined, and the meaning of the kana-written word is specified.

Further, in the invention according to claim 2, the input Japanese sentence is divided into a plurality of Japanese morphemes by the morphological analysis unit, and the morphological analysis is further performed to recognize the word and its part of speech. If there is a kana-written word in the divided word string,
A reading of the kana-written word is searched from the written information storage unit, and a plurality of kanji notations associated with the searched reading of the word are displayed on the display unit. Then, the kanji notation determination unit is operated by the operator, and a desired kanji notation is instructed and determined from among the plurality of kanji notations displayed on the display unit. Then, based on the determined meaning of the kanji notation and the analysis result of the morphological analysis unit, the syntax analysis unit analyzes the syntax of the input Japanese sentence. Based on the syntax analysis result, the translation unit performs The input Japanese sentence is translated into another language sentence. In this way, the meaning of the kana word in the input Japanese sentence is specified, and the language translation is performed efficiently and correctly.

In the invention according to claim 3, when the plurality of kanji notations retrieved from the notation information storage unit are displayed on the display unit by the notation information processing unit, the plurality of kanji notations are added to the kanji notation. Kanji notation with high kana writing frequency is displayed with priority. Therefore, the desired kanji notation is displayed without the kanji notation determination unit being repeatedly operated by the operator.

Further, in the invention according to claim 4, the Japanese sentence input from the input means is divided into a plurality of Japanese morphemes by the processing means, and morphological analysis is performed to identify the word and its part of speech. . If a kana word is present in the divided word string, a reading of the kana word is searched from among the readings of the Japanese words registered in the memory means. A plurality of kanji notations associated with the reading are displayed on the display means. Then, the operator selects and indicates a desired kanji notation from the displayed plurality of kanji notations.
Then, the meaning of the kana written word is specified based on the designated kanji notation.

Further, in the invention according to claim 5, the Japanese sentence input from the input means is divided into a plurality of Japanese morphemes by the processing means, and morphological analysis is performed to identify the word and its part of speech. . If a kana word is present in the divided word string, a reading of the kana word is searched from among the readings of the Japanese words registered in the memory means. A plurality of kanji notations associated with the reading are displayed on the display means. Then, the operator selects and indicates a desired kanji notation from the displayed plurality of kanji notations.
Then, the meaning of the kana written word is specified based on the designated kanji notation. After that, syntax analysis of the input Japanese sentence is performed based on the specified meaning of the kanji notation and the analysis result of the morphological analysis unit. Then, based on the result of the parsing, the input Japanese sentence is language-translated into another language sentence.

In the invention according to claim 6, when the plurality of kanji notations are displayed, the added kanji notation with a high kana writing frequency is preferentially displayed. Thus,
The desired kanji notation is specified early.

[0017]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the illustrated embodiments. In the present embodiment, the present invention will be described by taking as an example a case where the language processing apparatus of the present invention is applied to a Japanese-English machine translation apparatus. FIG. 1 is a schematic block diagram of a Japanese-to-English machine translation apparatus according to the present embodiment. Reference numeral 1 denotes a CPU (central processing unit), 2 denotes a display unit including a CRT (cathode ray tube), and 3 denotes an OCR (optical type). An input unit composed of a character reading device) and a keyboard; 4 is a main memory for storing translation programs and grammatical rules; 5 is a translation module for performing Japanese-English translation processing; 6 is a translation processing dictionary and notations described in detail below; The external storage unit stores information and the like. The Japanese sentence read by the OCR constituting the input unit 3 is
Under the control of the CPU 1, the translation module 5 refers to the translation processing dictionary and the notation information stored in the external storage unit 6, and is translated into English. FIG. 2 is a more detailed block diagram of the translation module 5. The dictionary lookup / morphological analysis unit 11 divides the input Japanese sentence as a source language into Japanese morpheme (word) strings,
The morphological analysis is performed by obtaining the morphological information (to be described later in detail) of each of the divided Japanese morphemes by looking up the translation processing dictionary stored in the translation processing dictionary stored in the Japanese dictionary. The notation information processing unit 12 displays, on the display unit 2, a plurality of kanji notations of the kana written Japanese words when the kana written Japanese words are present in the divided Japanese word strings. . Then, based on the key input from the input unit 3, the kanji representation of the Japanese word written in the kana is specified. The syntax analyzer 13 analyzes the syntax of the Japanese word string obtained from the input sentence with reference to the grammar rules stored in the main memory 4 and the obtained morpheme information. The conversion unit 14 converts the obtained syntax analysis result (syntax parse tree) in Japanese into a syntax parse tree in English with reference to the morphological information. The translated sentence generation unit 15 refers to the morphological information, corrects the obtained translated sentence based on the parse tree in English into a correct sentence as English, and generates a translated sentence in English. The generated translation is displayed on the CRT constituting the display unit 2.

Hereinafter, the translation dictionary stored in the external storage unit 6 will be described in more detail. FIG. 3 shows the morphological information as the contents of the translation dictionary. This morpheme information is composed of five pieces of information, ie, Japanese headline information 16, Japanese morpheme information 17, Japanese analysis information 18, Japanese-English conversion information 19, and English generation information 20. Usually, the above five morpheme information are provided for each word. Information is associated. Here, the Japanese morpheme information 17 is part of speech information, connection information, and the like of the Japanese morpheme, and is used in the above-described morphological analysis and syntax analysis. The Japanese analysis information 18 is semantic feature information of a Japanese morpheme, and is used at the time of semantic analysis.
The Japanese-English conversion information 19 is position information and the like that can be taken in an English sentence when a Japanese word is converted into an English word, and is used in the above-described conversion of the parse tree and the like. The English generation information 20 is translated word information of Japanese words and the like, and is used when generating the above-described translation.

Next, the notation information stored in the external storage device 6 will be described. This notation information is information used when specifying the kanji notation and meaning of Japanese words written in kana. FIG. 4 shows an example of the contents of the notation information. This notation information includes the Japanese word “Yomi (ie, Kana notation)” and the Japanese word “Japanese heading” in Kanji notation having the same reading as this Japanese word “Yomi”.
(That is, kanji notation) "and the Japanese word" sense "of the kanji notation having the same reading. Therefore, by referring to this notation information, the Japanese word having the same" reading "can be obtained. It is possible to know the kanji notation of all words and the meaning of each kanji notation.

That is, the external storage device 6 constitutes the notation information storage unit, and the input unit 3 constitutes the kanji notation determination unit.

The Japanese-English machine translator having the above configuration operates as follows. FIG. 5 is a flowchart of the Japanese-English translation processing operation executed by the translation module 5 under the control of the CPU 1. Hereinafter, the Japanese-English translation processing operation will be described in detail with reference to FIGS. In step S1, a Japanese sentence to be translated is input from the OCR constituting the input unit 3. In step S2, the dictionary lookup / morphological analysis unit 11
As a result, the input Japanese sentence is divided into Japanese morphemes. Further, based on the Japanese index information 16 of the translation processing dictionary stored in the external storage unit 6, each of the divided Japanese morphemes is searched. Then, Japanese morpheme information 17, Japanese analysis information 18, Japanese-English conversion information 19, and English generation information 20 of each Japanese morpheme are obtained. In step S3, a morphological analysis is performed by the dictionary lookup / morphological analysis unit 11 based on the search result in step S2 and the obtained Japanese morphological information 17. That is, the word recognition of each Japanese morpheme of the input Japanese sentence is performed, and the Japanese word and its part of speech are obtained.

In step S4, if there is a kana-written word in the Japanese word string obtained in step S3, the notation information processing unit 12 applies the following to the kana-written word. Notation information processing is performed. That is, the kana word is searched based on the “reading” column of the written information stored in the external storage unit 6 (see FIG. 4). As a result, if the same reading as the kana-written word is present in the “reading” column of the written information, the kana-written word has a plurality of “Japanese headlines (kanji notation)”. In other words, the kana word has a plurality of meanings and the meaning is ambiguous. Then, all the "Japanese headlines" associated with the "reading" of the retrieved notation information are displayed on the display unit 2, and the kanji of the kana-written word is displayed by an interactive process described later in detail. The notation is specified. At that time, the specified kanji notation is associated with the meaning as shown in FIG. Therefore, by specifying the kanji notation of the kana written word, the meaning of the kana written word is automatically specified.

In step S5, the grammatical rule stored in the main memory 4 and the Japanese language obtained as a result of the search are obtained for the word recognition result in which the meaning of the kana written word is specified by the notation information processing in step S4. Based on the word morpheme information 17, the Japanese analysis information 18, and the like, the syntactic analysis unit 13 performs syntactic analysis with semantic analysis. In this manner, the result of segmentation in the Japanese word string and the dependency between the segments are determined, and the parse tree is determined. In step S6, the Japanese parse tree obtained in step S5 is converted into an English parse tree based on the Japanese-English conversion information 19 obtained as a result of the search. In step S7, a correct translation is generated based on the English syntax parse tree obtained in step S6, using the English generation information 20 obtained as a result of the search. In step S8, the English translation sentence generated in step S7 is displayed on the CRT constituting the display unit 2, and the Japanese-English translation processing operation ends.

The above-mentioned Japanese-English translation processing operation will be described below with reference to FIGS. 1 to 10 by taking as an example a case where an input sentence "was here before" is translated into English as a target language. This will be described more specifically. The Japanese sentence read by the OCR is stored in the work area of the main memory 4 as shown in FIG. … Step S1 Then, for the Japanese sentence thus read,
Dictionary lookup / morphological analysis is performed using the translation processing dictionary. That is, the read Japanese sentence is divided into Japanese morpheme strings, the Japanese heading of the word is searched using the dictionary for translation processing, and the obtained Japanese heading information 1 is obtained.
The contents of the Japanese morpheme information 17, Japanese analysis information 18, Japanese-English conversion information 19, and English generation information 20 associated with 6 are stored in the buffer of the main memory 4. next,
Based on the contents of the Japanese morpheme information 17 stored in the buffer, a word recognition result as shown in FIG. 7 is obtained. … Step S2, Step S3

Next, based on the word recognition result shown in FIG. 7, notation information processing is performed using the display information. That is, the verb “Kake” in the word recognition result
Is a pseudonym. Therefore, the final form of the kana word "kake" is searched for based on the "reading" column of the written information. As a result, as shown in FIG. 4, “reading” and “coming” are divided into five “Japanese headlines” “(1) hanging over”, “(2) hanging over”, “(3) hanging over”, It can be seen that "(4) is affected" and "(5) is affected". Therefore, as shown in FIG. 8, the contents of the specific "Japanese headline" of the kana word "Kake" on the CRT display screen of the display unit 2 "Kakeru Kakeru", "Kakeru Kakeru", "Kakeru Kakeru" , "I get affected" and "I get involved" are displayed to urge the operator to specify the kanji representation of the kana word "Kake." At this time, the input sentence “was previously hung here” is simultaneously displayed on the display screen, and the place of the kana word “kake” in the input sentence is surrounded by a rectangle 21 so that the current window 22 Are displayed so that the operator can recognize the kana written word having a plurality of "Japanese headlines" and the Japanese sentence using the kana written word. Then, the operator selects, from among the plurality of “Japanese headings” displayed in the window 22, the “Japanese heading” most suitable as the kanji notation of the kana word “Kake” in the displayed input sentence.
Is indicated by the cursor 23 and the execution key or the like is depressed to determine. As a result, the kana word "kake" in the input sentence "was here before"
"Kakeru Kakeru" is specified as the most appropriate "Japanese headline" for. … Step S4

Next, a syntax analysis is performed based on the word recognition result and the contents of the Japanese morpheme information 17 and the contents of the Japanese analysis information 18 stored in the buffer.
First, using the grammatical rules stored in the main memory 4, the input Japanese sentence is divided into phrases based on the contents of the Japanese morphological information 17 and the word recognition results. The grammatical rule used at that time is, for example, a grammatical rule in which the left side of an arrow is a "phrase" and the right side is a "part of speech", for example, a noun phrase for continuous use → noun + case particle for continuous use. Then, the grammar rule as described above is applied to the word recognition result shown in FIG. 7 to divide the input word string into phrases. That is, for example, "here (pronoun)" in the word recognition result of FIG.
In the case of the word chain "-(particle)", the grammatical rule "Noun phrase for continuous use → Noun + case particle for continuous use" applies, so "here (pronoun) (particle)" is divided into one phrase. It is. In this way, a segmentation result as shown in FIG. 9 is obtained.

Further, using the above grammatical rules, the dependency between phrases is examined based on the result of segmentation. The grammatical rules used at that time are, for example, verb phrase → noun phrase for continuous use + verb phrase (application example 1: buy a book → buy a book +) (application example 2: I buy a book → I am a + book Like the grammar rule, the left side of the arrow is a "phrase" and the right side is a "phrase". Then, the grammatical rules as described above are applied to the phrase segmentation results shown in FIG. 9 to determine the interdependence between the segmented phrases. That is, for example, in the case of the phrase “here (pronoun) is (adjective)” and the phrase “kake (verb) is (auxiliary verb) ta (auxiliary verb)” in the phrase segmentation result of FIG. 9, the grammatical rule “verb phrase → Since the phrase "noun phrase + verb phrase" is applied, the phrase "here (pronoun) to (particle)" is replaced by the phrase "kake (verb).
(Auxiliary verb). At this time, the semantic analysis is executed using the semantic feature information and the case pattern of the verb "to be", which are the contents of the Japanese language analysis information 18, and the case pattern does not apply (in other words, Japanese) It is difficult to say that it is semantically invalid). As a result, a dependency (a parse tree) between the phrases as shown in FIG. 10 is obtained.

That is, in this embodiment, prior to performing the syntax analysis, the kanji mark of the kana-written word including the ambiguity of the meaning is interactively specified by the notation information processing. As a result, the kanji notation of the kana written word “Kake” of the input sentence “was here before” was specified as “Kakeru”, and the meaning of “Kakeru” in the meaning shown in FIG. There must be something to cross the object. " In this way, by defining the meaning of the verb "to be" unambiguously, the case pattern of the verb "to be" is uniquely obtained in the above-described semantic analysis. Therefore,
Multiple parse trees obtained based on the above grammar rules are
It is possible to narrow down to a more correct parse tree using a uniquely defined case pattern. On the other hand, when notation information processing is not performed on the input sentence, the following is performed. That is, since the meaning of the verb "take" has not been unambiguously determined prior to the parsing,
In some cases, a plurality of case patterns of the verb "take" may be obtained.
In that case, the dependency relationship between the clauses obtained based on the grammar rules cannot be narrowed down, and a parse tree that is not semantically valid is also generated. That is, in the present embodiment, by specifying the kanji notation of the kana-written word by the notation information processing, the meaning of the kana-written word including the ambiguity of the meaning is limited,
It increases the accuracy of parsing. … Step S5

The interdependency between phrases (ie, a Japanese parse tree) as shown in FIG. 10 obtained as described above is converted into an English parse tree. Then, an English translation is generated based on the obtained English syntax analysis tree, and the translation result is displayed on the CRT of the display unit 2. At this time, since the correct meaning of the kana word "Katsu" is determined by the interactive form via the CRT as described above, a translated sentence that correctly reflects the intention of the operator is generated. However, if the meaning of the pseudonym word "take" is not specified, a mistranslation that does not reflect the contents of the text is output, and the labor of post-editing increases. ... Steps S6 to S8

As described above, in the present embodiment, in the external storage unit 6, “Japanese heading (Kanji notation)”, “Yomi (Kana notation)”, and “Word meaning” of a plurality of words having the same kana notation "
Is stored. Then, at the time of the translation process, after the input Japanese sentence is subjected to dictionary lookup / morpheme analysis by the dictionary lookup / morphological analysis unit 11, the notation information processing unit 12 performs the following before the syntax analysis by the syntax analysis unit 13. Such notation information processing is performed. That is, if the word recognition result obtained as a result of the morphological analysis includes a kana word that is not in the Japanese index information 16 of the translation processing dictionary stored in the external storage unit 6, the above described notation information “ The kana-written word is searched for based on the "reading" column. If the same content as the reading of the kana-written word is present in the “reading” column, all “Japanese headlines” associated with the “reading” are displayed on the display unit 2 together with the input sentence. indicate. Then, the operator operates the display unit 2
With reference to the input sentence indicated in (1), "Japanese headline" most suitable as the kanji notation of the kana-written word in the input sentence is specified after operating the keyboard constituting the input unit 3 and instructing. Thus, by specifying the kanji notation of the kana-written word, the meaning of the kana-written word having a highly ambiguous meaning is specified. Therefore, according to the present embodiment, the ambiguity of the meaning of the kana-written word in Japanese is removed, and efficient Japanese-English translation processing becomes possible.

FIG. 11 is a flowchart of a Japanese-to-English translation processing operation of the Japanese-to-English machine translation apparatus in the above embodiment, which is different from FIG. The difference between this flowchart and the flowchart shown in FIG. 5 according to the above-described embodiment lies in the position of the steps of the notation information processing. That is, in the above-described embodiment, the notation information processing is performed prior to the syntax analysis. However, in the present embodiment, the notation information processing unit 12 shown in FIG. 2 is arranged between the syntax analysis unit 13 and the conversion unit 14, and the notation information processing is performed after the syntax analysis. In the case of the present embodiment, since the syntax analysis is performed before the notation information processing is performed, the kanji notation and the meaning of the kana-written word in the input sentence cannot be specified at the time of the syntax analysis. Therefore, the kana word in the input sentence is subjected to syntax analysis with the kana notation. As a result, as described in the above embodiment, if there is a kana word in which the meaning cannot be specified, the case pattern is not uniquely determined, so that a plurality of parse trees are obtained. After that, the above-described notation information processing is performed to specify the kanji notation of the kana-written word whose meaning cannot be specified, determine the meaning of the kana-written word, and narrow down the parse tree.
Therefore, the ambiguity of the meaning of the kana word in Japanese is removed, and efficient language processing can be performed. As described above, the above described notation information processing exhibits its effect regardless of whether it is executed before or after the syntax analysis. Therefore, the notation information processing may be performed at a stage convenient for the user during the translation process.

In each of the above embodiments, the order of the "Japanese heading" displayed on the display unit 2 in the notation information processing is the same as the order of the "Japanese heading" of the notation information. Therefore, if the kanji notation of a word that is frequently written in kana is stored below the column of “Japanese heading” in the notation information, the notation information processing related to the word that is frequently written in kana is performed by the operator. In this case, it is necessary to repeatedly operate the cursor to specify the "Japanese headline" displayed at the bottom of the window 22, which is troublesome. Therefore, in the present embodiment, the kana name frequency is added in advance to the “Japanese headline” in the above described notation information for each “Japanese headline”. In other words, “Japanese headlines” of words that are likely to be written in kana, for example, difficult characters… “oysters” names of animals and plants… “persimmons” and “dogs” functional verbs… Accordingly, information “many” or “slightly many” representing the kana name frequency is added, and information “small” representing the kana name frequency is added to the “Japanese headline” of other words. As a result, for example, as shown in FIG. 12, three ranks of the pseudonym frequency are added to the “Japanese headline” of the notation information related to the kana notation “Kaki”. Then, when executing the notation information processing, as shown in FIG. 13, the “Japanese headline” of the kana written word is displayed on the display unit 2. That is, the input sentence “I bought kaki yesterday.” Is displayed on the display screen at the same time, and the portion of the kana word “Kaki” in the input sentence is surrounded by the rectangle 31. When a plurality of dictionary headings are displayed in the window 32, “Japanese headlines” having the highest rank of the added kanabook frequency are sequentially displayed. The headlines are displayed in order from the “Japanese headlines” stored at the top of the column. At this time, the initial position of the cursor 33 is set to the position of the highest "Japanese headline" displayed in the window 32. As described above, in the present embodiment, when performing the notation information processing, the “Japanese headline” having a high pseudonym frequency is displayed above the window 32. Therefore, when specifying the “Japanese heading” most suitable as the kanji notation of the kana word “Kaki” displayed in the input sentence by the rectangle 31 from the “Japanese heading” displayed in the window 32, The number of operations for moving is reduced, and the burden on the operator is reduced. At this time, when adding the kana name frequency to the “Japanese heading”, the kana name frequency is added to each “Japanese heading” in advance based on a certain criterion, and the operator can add the kana name frequency as necessary. If the content of the added kana letter frequency is changed, the burden on the operator can be further reduced.

In the above embodiment, a kana name frequency is added in advance to each “Japanese headline” in the notation information,
“Japanese headlines” with the highest rank of added kana-semigraphy
, And if the rank of the kana name frequency is the same, the “Japanese headline” is displayed in the window 32 in order from the “Japanese headline” stored in the higher rank. However, in the above display information processing, it is natural that the “Japanese headline” having a high rank of the added kana name frequency is specified by key input from the input unit 3 in many cases. Multiple "Japanese headlines"
Need to be displayed in the window 32 to request the operator to specify the "Japanese headline". Therefore, at the time of the notation information processing, the “Japanese headline” having the highest kanagraph frequency may be automatically applied as the kanji notation of the kanawritten word. If an appropriate parse tree cannot be obtained as a result of performing the parsing based on the fitted “Japanese heading”, the “Japanese heading” having the next kana book frequency is automatically added to the kana. This is the kanji notation of the written word. By doing so, the operator can be relieved from the inconvenience of "Japanese headlines".

In the notation information processing in each of the above embodiments, the display format of the "Japanese headline" relating to the target kana written word is not limited to FIGS. In each of the above embodiments, the case where the language processing apparatus of the present invention is applied to a Japanese-English machine translation apparatus has been described.
It is not limited to this. That is, the present invention can be applied to a Japanese summarizer mainly for processing text written in Japanese, a computer using a natural language interface in Japanese, a robot controller, and the like.

[0035]

As is clear from the above description, the language processing apparatus according to the first aspect of the present invention divides an input Japanese sentence into a plurality of Japanese morphemes by a morphological analysis unit, performs morphological analysis, and performs word processing. And the part-of-speech are recognized, and if the kana word is present in the divided word string, the notation information processing unit associates the kana word with the reading of the kana word searched from the notation information storage unit. A plurality of kanji notations are displayed on a display unit, and a kanji notation determining unit operated by an operator designates and determines a desired kanji notation from the plurality of displayed kanji notations. Therefore, the kanji notation of the kana written word is determined, and the meaning of the kana written word is specified. That is, according to the present invention, the ambiguity of the meaning of the kana word in the Japanese text is reduced, and efficient language processing can be performed.

Further, in the language processing apparatus according to the second aspect of the invention, the morphological analysis unit divides the input Japanese sentence into a plurality of Japanese morphemes, performs morphological analysis, and recognizes words and their parts of speech. If there is a kana word in the divided word string, the notation information processing unit displays a plurality of kanji notations associated with the reading of the kana word searched from the notation information storage unit. Is displayed, and a desired kanji notation is designated and determined from among the plurality of displayed kanji notations by a kanji notation determining unit operated by an operator. Further, based on the determined meaning of the kanji notation and the analysis result of the morphological analysis unit, a syntax analysis of the input Japanese sentence is performed by the syntax analysis unit. Based on the syntax analysis result, the translation unit performs the syntax analysis. The input Japanese sentence is translated into another language sentence. Therefore, when the kanji notation of the kana-written word is determined, its meaning is specified. That is, according to the present invention, the ambiguity of the meaning of the kana words in Japanese text is reduced, and efficient and correct language translation processing can be performed.

The language processing apparatus according to claim 3, when the plurality of kanji notations retrieved from the notation information storage unit are displayed on the display unit by the notation information processing unit,
Since the added kanji notation with high kana writing frequency is preferentially displayed, the operator can easily determine the kanji notation without operating the kanji notation determination unit more than necessary. Therefore, according to the present invention, the ambiguity of the meaning of the kana words in Japanese is easily reduced, and more efficient linguistic processing becomes possible.

According to a fourth aspect of the present invention, there is provided a language processing method, wherein a Japanese sentence input from an input unit is divided into a plurality of Japanese morphemes by a processing unit, and morphological analysis is performed to convert a word and its part of speech. If there is a kana word in the divided word string, the kana word is read from the reading of the Japanese word registered in the memory means. A plurality of kanji notations associated with the reading of the word are displayed on the display means, a desired kanji notation is selected from the displayed plurality of kanji notations, and the designated kanji is indicated. Since the meaning of the kana-written word is specified based on the notation, the ambiguity of the meaning of the kana-written word in the Japanese text can be reduced. Therefore, according to the present invention, efficient language processing can be performed when operating a Japanese-English machine translator, a Japanese summarizing machine, a Japanese interface, or the like.

According to a fifth aspect of the present invention, there is provided a language processing method, wherein a Japanese sentence input from an input unit is divided into a plurality of Japanese morphemes by a processing unit, and morphological analysis is performed to convert a word and its part of speech. If there is a kana word in the divided word string, the kana word is read from the reading of the Japanese word registered in the memory means. A plurality of kanji notations associated with the reading of the word are displayed on the display means, a desired kanji notation is selected from the plurality of displayed kanji notations, and the selected kanji notation is designated. The input Japanese sentence is subjected to syntax analysis based on the analysis result of the morphological analysis unit, and the input Japanese sentence is translated into another language sentence based on the syntax analysis result. of It can reduce the ambiguity of taste. Therefore, according to the present invention, efficient and accurate language translation processing can be performed.

Further, in the language processing method according to the present invention, when displaying the plurality of kanji notations, the added kanji notation with a high kana writing frequency is preferentially displayed. The desired kanji notation can be specified early. Therefore, according to the present invention, more efficient language processing can be performed.

[Brief description of the drawings]

FIG. 1 is a schematic block diagram of an embodiment of a Japanese-English machine translation apparatus to which a language processing apparatus according to the present invention is applied.

FIG. 2 is a detailed block diagram of a translation module in FIG. 1;

FIG. 3 is a diagram showing morphological information in a dictionary for translation processing.

FIG. 4 is a diagram showing an example of the content of notation information.

FIG. 5 is a flowchart of a Japanese-English translation processing operation by the Japanese-English machine translation device shown in FIGS. 1 and 2;

FIG. 6 is a diagram illustrating an example of an input sentence.

FIG. 7 is a diagram showing an example of a word recognition result for the input sentence shown in FIG. 6;

8 is a diagram showing a display example of a kana-written word and a plurality of Japanese headlines related to the kana-written word in the input sentence shown in FIG. 6;

9 is a diagram illustrating an example of a segmentation result of an input sentence based on the word recognition result illustrated in FIG. 7;

FIG. 10 is a diagram showing an example of dependency between phrases based on the phrase segmentation result shown in FIG. 9;

FIG. 11 is a flowchart of a Japanese-English translation processing operation different from FIG. 5;

FIG. 12 is a diagram illustrating an example of the content of written information to which a kana name frequency is added;

13 is a diagram illustrating a display example of a kana written word and a plurality of Japanese headlines related to the kana written word in the notation information processing based on the notation information shown in FIG.

[Description of Signs] 1 ... CPU, 2 ... Display unit, 3 ... Input unit, 4 ... Main memory, 5 ... Translation module, 6 ... External storage unit, 11 ... Dictionary / morphological analysis unit, 12 ... Notation information processing , 13 ... Syntax analysis unit, 14 ... Conversion unit, 15 ... Translation sentence generation unit, 16 ... Japanese headline information, 17 ... Japanese morpheme information, 18 ... Japanese analysis information, 19 ... Japanese-English conversion information, 20 ... English generation information, 22, 32 ... window, 23, 33 ... cursor.

Continued on the front page (72) Inventor Mari Tokoda 22-22 Nagaikecho, Abeno-ku, Osaka-shi, Osaka Inside Sharp Corporation (56) References JP-A 1-286075 (JP, A) JP-A 63-276174 (JP, a) JP flat 2-25967 (JP, a) JP Akira 63-219061 (JP, a) JP Akira 62-175860 (JP, a) (58 ) investigated the field (Int.Cl. ⁷ G06F 17/20-17/28

Claims (1)

(57) [Claims] An input unit for inputting a Japanese sentence; a display unit for displaying a Japanese sentence input by the input unit; and a Japanese sentence input by the input unit is divided into a plurality of Japanese morphemes. A morphological analysis unit that analyzes and recognizes the word and the part of speech of the word, reading of the word, a plurality of kanji notations having the reading, and the meaning of each kanji notation are stored in association with each other. A notation information storage unit, when there is a kana written word in a word string obtained from the input Japanese sentence obtained by being analyzed by the morphological analysis unit, And a notation information processing unit for displaying on the display unit a plurality of kanji notations associated with the searched reading of the kana word, and a plurality of notations displayed by the notation information processing unit Kanji of A kanji notation determining unit for designating and determining a desired kanji notation from among the kanji notations, based on the kanji notation of the kana written word determined by the kanji notation determining unit. The kana writing frequency is added to each of the kanji notations stored in the notation information storage unit, and the notation information processing unit is configured to display the plurality of kanji notations. Is a language processing apparatus characterized in that the added kanji notation with a high kana writing frequency is preferentially displayed.