JPS60193073A - Japanese word analyzer - Google Patents

Abstract

PURPOSE:To decide well-definedly the grammatical structure of an input sentense by displaying the input sentences, related word candidates and meaning candidates succcessively to a display means and analyzing the input sentence in response to an indication given from an input means. CONSTITUTION:A sentence structure analyzer 30 retrieves the grammar of a file 42 and analyzes the sentence structure. A controller 20 extracts a word number 1 out of a memory 22 in response to the end of the sentence structure analysis. Then a related word candidate output device 26 transfers and stores the related word candidates to a related word buffer 44; while a meaning candidate output device 28 transfers and stores the meaning candidates to a meaning candidate buffer 46. The controller 20 give indications to both devices 26 and 28 to read the storage contents of both buffers 44 and 46. An operator checks his/her desired operation while looking at a display.

Description

[Detailed Description of the Invention] The present invention relates to a Japanese language analysis device, more specifically, a translation machine,
The present invention relates to a Japanese language analysis device that is applied to processing systems such as word processors, office computers, and personal computers, and in which the user base intervenes.

Traditionally, in Japanese analysis in machine translation, for example,
Broadly speaking, the following three input methods were used. One method is to input the input sentence as it is without making any modifications. This is because a plurality of analysis results are obtained, and it is not determined uniquely, and it is not always possible to obtain the results intended by the operator.

As a second input method, there is a method in which the input sentence is corrected and then input. This method also has the same drawbacks as the first method, but the number of analysis results obtained is smaller than the first method.5 Moreover, the probability of obtaining correct results is high.

Further, as a third method, there is a method in which the input sentence is limited to a specific format. This has the advantage that the analysis result is uniquely determined, but if the operator does not adhere to the limitations, the correct answer may not be obtained.

In particular, the second and third methods require specialized knowledge to modify the input sentence into a predetermined format, and therefore have difficulties in the operability and usability of the devices.

If an attempt is made to realize a Japanese language analysis device that solves these problems, the Japanese grammar analysis program will become extremely complex and require a great deal of effort to create.

However, it is an object of the present invention to overcome the drawbacks of the prior art and to provide a Japanese language analysis device that can uniquely determine the grammatical structure of an input sentence intended by the user.

The structure of the present invention will be described below with reference to Example 1.

Referring to FIG. 1, the embodiment of the Japanese language analysis device according to the present invention includes a printer 10 as a print output device, a display 12 such as a CRT for visually displaying characters, and a keyboard 14 for inputting Japanese sentences using kana or the like. and these are connected to the control device 20. In addition to the kana key, the keyboard 14 also has function keys such as a "next candidate key" and "OK primary key.The next candidate key is used to select the next modification candidate for a modifier or the meaning candidate for a word, as described below. This is an instruction key to shift to the item with the rank of
The primary key is an instruction key that is operated when one of the candidates matches the intended one to move to the next operation step.

The control device 20 is implemented as a processing system, and controls various parts of the Japanese language analysis device including display screen control of the display 12. Recorded as the system memory of the control device 20 tα
A cooking tank 22 is provided. This is used to temporarily store the processing results of the language analysis device +y, +, and is used as a storage area for other programs and data.

A morphological analyzer 24, a syntactic analyzer 30, a dependency candidate output device 26, and a semantic candidate output device 28 are controlled by the control device 20 and execute the analysis operations at the end of the Japanese language analyzer. These devices 24, 30, 26 and 28 may be included in the processing system of the control device 20 or may be realized in a separate processing system.

Connected to the morphological analysis device 24 is a file storage device 40 such as a magnetic disk that stores a connection table and a word dictionary that define connections between words. The word dictionary is
The notation, pronunciation, part of speech, meaning, attributes, etc. of words are described for each unit. The morphological analysis device 24 is a device that indexes the file 40, decomposes a Japanese input sentence inputted in the form of, for example, a kana kanji situation sentence into word units, and analyzes the eutectic part of speech.

Connected to the parsing device 30 is a file writing device 42 in which a grammar dictionary describing parsing grammars is stored. The J-data format is as shown in Section 212.
The growth and taste are listed accordingly.

An example of the parsing grammar stored in this grammar dictionary is partially shown in diagram ft52A. What kind of question is α in 14?
This indicates that it may be 5/i. In the following description and illustrations, in order to avoid complicating the description, the attributes of words will be omitted since they are not directly relevant to the understanding of the present invention.

As shown, the parser 2) 30 consists of a control section 32 and a pair of memory stacks 34A and 34B. The syntactic analysis device 30 is a device that indexes the grammar dictionary 42 and analyzes the input Japanese sentence, which has been decomposed into parts of speech by the morphological analysis device 24, for the syntactic structure of each word. In this embodiment, the grammatical structure of the analysis result is unique, and ambiguity is output linked to each word. Furthermore, their priority is
Depending on the conditions of each word, the dependency relationship of the closer one becomes higher.

Dependency candidate output device; ξ26 includes dependency buffer 4
4 is connected, and all words (modified words) that can be related to the word of interest, that is, the word being processed, are stored in this number. In the stored data format, as shown in FIG. 3, the modified word numbers are listed in order from the highest priority to the lowest in the order of arrow A. The modification candidate output 'A position 26 is a device that selects modification candidates from the modification buffer 4B.

A meaning buffer 46 is connected to the meaning candidate output device 28, and the latter stores all meanings of the word of interest. As shown in FIG. 4, the accumulated data format is such that the meanings of each word are listed in the order of arrow B, from highest to lowest priority, corresponding to the number of the modified word. Access this and output semantic candidates! A position 28 selects meaning candidates for the modifier.

The data as a result of analysis by these devices is temporarily stored in the storage device 22, and the data format of the syntax analysis results is shown in FIG. 5, and the final result obtained by this device is shown in FIG. The data format is shown in . Note that a plurality of numbers may be included in the storage slot that stores the number of each modifier.

The operation of this device will be explained. For example, suppose that the sentence ``tall marble desk'' is input from the keyboard 14. The intended meaning is an expensive desk made of marble;
Assume that the height of the marble is not high.

This input sentence is temporarily stored in the recording device 22 by the control device 20 and is visually displayed on the display 12. At the same time, the morphological analysis device 24 morphologically analyzes this input sentence. This may be done using the connection table/word dictionary 40 in a conventional manner. The analysis results are shown in FIG. This is transferred to the parser 30,
They are accumulated in stacks 34A and 34B, respectively, as shown in FIGS. 8A and 8B. That is, the currently focused word (modifier) is stored in the stack 34A, and the modified word of the word in the stack 34A is stored in the stack 34B.

Therefore, the syntax analysis device 30 stores the file 4 shown in FIG. 2A.
2 is indexed and parsed. First, depending on each state of the stack 34A, in this example, the (
1) to (8) are checked. In this example, for (1) and (5), there is no candidate in the stack 34B that corresponds to "noun + α" and 1 person, so reject I
・To be done. Other (2) to (4) (e) to (8)
Since there is a corresponding candidate in the stack 34B, the analysis result shown in FIG. 5A is obtained. This is expressed in the data format shown in FIG.

Next, the syntax analyzer 30 shifts word number 2 "Marble J" from stack 34B to stack 34A, and word number 3 "Desk J" remains in stack 34B. and obtain the syntax analysis result.

Similarly, 1t/! The M number 3 "desk" is transferred from the stack 34A to the stack 348, but in this example, since this is the end of the sentence, the analysis ends here. Therefore, the fifth
The syntax analysis result shown in Figure B is obtained. This is control device 2
The data is transferred from 0 to the storage device 22 and stored. Note that this syntax analysis result data is written in the format shown in FIG.

The control device 20 stores the storage device 22 in response to the completion of the parsing.
Extract the word number l from. Therefore, the modification candidate output device 26 outputs the modification candidate to the modification buffer 44.
Then, the meaning candidate output device 46 transfers the meaning and flavor candidates to the meaning candidate buffer 46 and stores them therein. Its condition 1
1; is shown in FIG. 3A.

The input sentence has been displayed on the display 12 so far, but the column word, that is, the word number l "high" is displayed in reverse video in this embodiment. This is shown in the hatched area 52 on the display screen 50 of the display 12 in FIG. 9A.
Indicated by

Therefore, the control device 20 sends a signal requesting a modification candidate to the modification candidate output device 2B. In response to this, the modification candidate output device 26 indexes the modifications, retrieves the word with the highest priority, that is, the modified word number 2 at the left end in FIG. 3A, and transfers it to the control device 20. Control device 20
blinks the display word "marble" corresponding to word number 2. This is indicated by a rectangular dotted line 54 in FIG. 9B.

The operator then checks whether the word in the link is the intended dependency word or not, and since it is not the intended word in this example, he operates the next candidate key on the keyboard 14. In response to this, the control device 20 requests the dependency output candidate device 26 for the next highest priority candidate in the same manner as described above. As a result, the next ranked modified word number 3 is retrieved from the dependency buffer 44 and transferred to the control device 20. Then, the control device 20 stops the blinking of the displayed "marble" and causes the "desk" to be printed.

The operator again checks whether the word in the link is the intended word or not, and since it is the intended dependent word in this example, he operates the OK primary key on the keyboard 14. In response to this, the control device 20 stops the display blinking of desk j, sends a signal indicating candidate confirmation to the modification candidate output device 26, and the device 26 outputs the modified word number 2 from the modification buffer 44. Erase.

Next, the control device 20 sends a signal to the meaning candidate output device 28 requesting the meaning candidate of the modifier number 1 for the modified word number 3. In response, the meaning candidate output device 28 indexes the meaning buffer 4B and retrieves a candidate with a high priority for the modified word number 3, that is, the meaning "evaluation" at the top in FIG. 3A. This is transferred to the control device 20 and is transmitted to the display area 56 of the display 12 as shown in FIG. 8C.
will be displayed.

The operator checks this display 5θ to see if this meaning is what he or she intended.In this example, since it is not what he intended, he operates the next candidate key on the keyboard 14. In response to this, the control device 20 requests the meaning candidate output device 28 for the candidate with the next highest priority in the same manner as described above. As a result, the meaning "value" of the next rank is retrieved from the meaning buffer 46 and transferred to the control device 20. Therefore, the control device 20 causes the display area 56 to display the information. This state is shown in FIG. 8D.

The operator checks the display 5G again to see if the taste, α, is the one he/she intended, and in this example, since it is the one he/she intended, he/she operates the OK primary key on the keyboard 14. In response to this, the control device 20 stops the inverted display 52 of the attention word "Kai" on the display screen 50 and erases the meaning display 56. At the same time, a signal indicating confirmation of the candidate is sent to the meaning candidate output device 28, and the device 28 deletes meanings other than the value J from the meaning buffer 46. This state is shown in FIG. 3B.

The control device 20 provides a dependency buffer 4 for the dependency candidate output device 26 and the meaning candidate output device 28, respectively.
4 and the contents stored in the meaning buffer 46 are read out. In addition, the notation and parts of speech (attributes) are read out from the syntax analysis results stored in the storage device 22, and
Organize the data as shown in the diagram.

In the same way, each subsequent word of the input sentence is processed and analyzed in this way, and when all words have been processed,
The analysis results are as shown in the 6th day diagram.

As described above, according to this embodiment, the response speed of the interaction process between the operator and the present device is fast, and syntax analysis does not need to be performed more than once.

Referring to FIG. 10, another embodiment of the invention is shown. In the subsequent figures, the same components as in FIG. 1 are indicated by the same reference numerals to avoid duplication of explanation.

The difference from the embodiment shown in FIG. 1 is that the dependency candidate output device 126 uses grammar to extract dependency candidates from the input sentence. Therefore, the modification buffer 44 in the embodiment shown in FIG. 1 is not provided in this embodiment, and the modification candidate output device 12B is connected to the meaning buffer 148.

In the meaning buffer 14B, as shown in FIG. 11, meaning candidates are accumulated in the direction of arrow C from higher priority to lower priority. The meaning candidate output device 128 is the meaning buffer 14
This is a device that selects meaning candidates from 6.

The syntax analysis device 130 is also connected to the modification candidate output device 126, and indexes the grammar dictionary 42 and performs analysis with priority given to the closest modification. Therefore, the analysis results are unambiguous and a single analysis result is output. The grammar dictionary 42 stores the grammar shown in FIG. 2B. Note that this shows part of the storage grammar, /(
slash) means "or".

The operation of the apparatus shown in FIG. 1O will be described using the same input sentence example as in the description of the embodiment shown in FIG. Input sentences stack 34A and 34
The process until the data is stored in B is the same as the embodiment shown in FIG. In other words, the contents are as shown in FIGS. 8A and 8B.

Depending on the state of stack 34A, grammar (1) (FIG. 2B) is applied to modified word number 2 of stack 34B.

Since this matches the grammar, the analysis for word number l is ended and the next word is shifted onto the stack 34A. The analysis result is as shown in FIG. 80. If even one grammatical analysis is successful, the analysis of that word ends and the process moves on to the analysis of the next word.

Subsequent words of this input sentence are analyzed in the same manner, and the analysis results shown in FIG. 6D are obtained. This is transferred from the control device 20 to the storage bag M22 and stored.

When the syntactic analysis is completed, the control device 20 first reads out the word number l, "high", from the storage device 22 and displays it in reverse video 52 on the screen 50 (FIG. 8A).

Next, the control device 20 checks the modified word section of word number 1, takes out the modified word number 2, and links it 54 on the display screen 50 (FIG. 8B).

Since this is different from the intended modification, the operator operates the next candidate key. As a result, the control device 20 activates the dependency candidate output device 128.

In response, the dependency candidate output device 126 outputs word number 2.
Using other words as candidates, perform grammatical analysis of word number l using Grammar Dictionary 4.
Do this by indexing 2. As a result, the analysis results shown in FIG. 6E are obtained. Among these, the meaning is transferred to the meaning buffer 14B and stored therein, and the meaning ztr is emptied and replaced with the previous syntax analysis result. This result is notified to the operator by displaying the "marble" in a blinking manner.

When the operator presses the OK main key, the control device 20 stops blinking and requests the meaning candidate output device 128 to provide a meaning candidate. The meaning candidate output device 12B is a meaning buffer 148
Extract "Evaluation" from the top of the page.

Transfer to the control device 20. This is displayed on the screen 50 in the state shown in FIG. 8C.

Then, when the operator presses the next candidate key, the control device 20 inputs the next candidate into the meaning candidate output device 128.

The next candidate is displayed as shown in Figure 8D.

When the operator presses the OK primary key, the controller will first
Stops highlighting and clears the meaning. Next, the analysis result is retrieved from the storage device 22 and the meaning term is "value".
Make it.

The results of the same process for all words are shown in 6B.
As shown in the figure.

In this way, in this embodiment, since there is no need to deal with ambiguity in syntax analysis, no unnecessary processing is performed, and the storage capacity of the processing system can be smaller than in the embodiment shown in FIG. It has characteristics.

Referring to FIG. 12, in yet another embodiment of the present invention, the grammar table 250 and semantic buffer 246 differ from the embodiment of FIG.
Grammars that are simpler or have weaker conditions have been accumulated. The data format is shown in FIG. The meaning buffer 24B temporarily stores meaning candidates, and its data format is the same as that shown in FIG. 11. The operation of the embodiment shown in FIG. 12 is basically the same as that shown in FIG. 10. The difference is that when parsing is performed in the state shown in FIG. 8A, the grammar dictionary 42 is used in the embodiment shown in FIG. 1θ.
However, in the embodiment of FIG. 12, the grammar table 250 is referred to instead. As a result, the analysis results shown in FIG. BE are obtained. Among these analysis results (FIG. 6E), the meaning is transferred to and stored in the meaning buffer 246, and the meaning term is emptied and replaced with the previous syntax analysis result. Thereafter, meaning candidates are selected from the meaning buffer 24B and displayed and processed in the same manner as in the embodiment shown in FIG.

As described above, the embodiment shown in FIG. 12 not only has the features of the embodiment shown in FIG. 1θ, but also has faster processing speed than the embodiment shown in FIG. There are symptoms.

Fuji-] According to the present invention, when a sentence is manually written, there is no need to correct it, simple grammar can be used, and the operator is not required to have specialized knowledge. In addition, the meaning and syntactic structure of the sentence intended by the operator can be uniquely obtained.
It is characterized by fast analysis speed. The present device is particularly effective for a method that uses a restricted grammar and inputs sentences with certain restrictions.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the Japanese language analysis device according to the present invention; FIG. 2, FIG. 2A, FIG. 2B, FIG. 3, FIG.
Figure B, Figure 4, Figure 5, Figure 5A, Figure 5B, Figure 6,
Figures 6A to 8E, Figure 7, Figure 8A, Figure 8B, and Figures 8A to 8D are explanatory diagrams for explaining the operation of the embodiment shown in Figure 1, and Figure 1θ is A block diagram showing another embodiment of the present invention, FIG. 11 is an explanatory diagram for explaining the operation of the embodiment shown in FIG. 10, FIG. 12 is a block diagram showing still another embodiment of the present invention, FIG. 13 is an explanatory diagram for explaining the operation of the embodiment shown in FIG. 12. 1:S1izu/) ,'p,H 3P... Display 14, Keyboard 20, Control device 22, Storage device 24, Morphological analysis device 28, Dependency candidate output Device 28, ... Meaning candidate output device 30, ... Syntactic analysis device 42, ... Grammar dictionary 44, ..., Dependency buffer 4B, ... Meaning buffer Patent applicant: Engraving of drawings by Ricoh Co., Ltd. (No change in content) Fig. 2 42 Fig. 2A 2 Fig. 2B Fig. 3 4 Fig. 4 6 Fig. 3A 2: Fig. 3B Fig. 5 Fig. 6 No' Fig. 5A Fig. 58 Fig. 6A Fig. 68 Fig. 6C Fig. 60 Fig. 6E Fig. 7 Fig. 8A Fig. 88 Fig. 9A Fig. 98 Fig. 90 @ Fig. 9D Procedural amendments April 11, 1981 Commissioner of the Japan Patent Office Kazuo Wakasugi 1. Display of 4 items 1959 Patent Application No. 47127 2, Name of Invention Japanese Language Analyzer 3, Relationship with Supplementary Person Case Patent Applicant Address 1-3-6 Nakamagome, Ota-ku, Tokyo Name (
874) Riko Co., Ltd. -4, Agent address: 1-13-4 Toranomon, Minato-ku, Tokyo 105 6. Contents of amendment The drawings submitted at the time of filing are replaced with official drawings (no changes to the engraving contents) as shown in the attached sheet. 7. List of attached documents (1) One official drawing

Claims (1)

[Scope of Claims] An input means for inputting an input sentence and an instruction in Japanese language, a morphological analysis means for analyzing morphemes included in the input sentence, and a parsing grammar for analyzing Japanese syntax, a syntactic analysis means for analyzing the syntax of the input sentence parsed into the morphemes using the grammar; and a dependency candidate selection means for selecting a dependency word candidate for a/i included in the input sentence parsed into the morphemes. , meaning candidate selection means for selecting meaning candidates for words included in the input sentence; storage means for holding the input sentence, the selected dependency candidates and meaning candidates, and the analysis results; and displaying Japanese. and a control means for controlling the input means, morphological analysis means, syntactic analysis means, dependency candidate selection means, meaning candidate selection means, storage means, and display means, and the control means controls the input sentence. , and the selected modification candidates and meaning candidates are sequentially displayed on the display means, and the input sentence is analyzed in response to an instruction manually inputted from the input means.