JPH06124157A - Word and phrase inputting device and method - Google Patents

Abstract

PURPOSE:To attain the input of a sentence which is correct in a structure and meaning without typing and inputting characters from a keyboard by offering the candidate of a word and phrase which can be connected just behind and after a certain word and phrase by using constraint information related with a syntax or the constraint information related with the meaning of the word and phrase. CONSTITUTION:When a word inputting instruction is selected (step 1), a word candidate which is correct in the structure and meaning is estimated (step 2) by referring to a word dictionary 11, syntax and meaning definition table 12 in which the constraint information related with the syntax and meaning which can be accepted as an input sentence is defined, and word table 13 inputted by a user. The word candidate as the estimated result is displayed to the user (step 3). Next, a word establishing instruction is selected, and the input of the word is established (step 4). At that time, the dictionary information of the established word and the information related with the syntax and meaning definition table 12 is stored in the word table 13, and held (step 6).

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information processing system in which a phrase such as a natural language is input, and more particularly to a phrase input device and method effective when there is a clear restriction on the connection relation of the phrase.

[0002]

2. Description of the Related Art Up to now, in many information processing systems that input a phrase such as a natural language, a keyboard is often used to input the phrase. Keyboard input is convenient for users who are accustomed to using the keyboard, but for users who are not accustomed to using the keyboard, it takes time to input, making the system unusable.

As one of means for solving this problem, as described in Japanese Patent Laid-Open No. 3-102564, when inputting a Japanese sentence, a syntax which the system can accept as an input sentence is held in advance as a template. However, there is a technique in which an input sentence is created by presenting it when the user inputs it and filling in the variable part.

[0004]

In the above-mentioned conventional technique, a new template needs to be prepared each time when the degree of freedom of the input sentence is increased by changing the attached word or inserting a new word. Also, when creating the input, the user has to select an appropriate template from among many templates. In addition, when filling in the variable parts of the template, it is necessary to input a phrase with a keyboard, which is time-consuming for a user who is not good at the keyboard type.

An object of the present invention is to provide a word / phrase input device which can input a natural language without a keyboard and can endure an increase in the flexibility of an input sentence.

[0006]

In order to achieve the above object, the present invention has knowledge defining constraints on the syntax of an input sentence and constraints on the meanings of words constituting an input sentence, and refers to this knowledge and a dictionary. Thus, the phrase input device is provided with means for estimating a phrase that can be input and displaying the phrase extracted by the estimation as a candidate.

[0007]

[Operation] By using the constraint information regarding the syntax and the constraint information regarding the meaning of the phrase, a candidate of connectable phrase is presented immediately after (immediately before) a certain phrase (including the case where no word is input yet). , You can input syntactically and semantically correct sentences without typing characters from the keyboard. In addition, it is possible to flexibly deal with an increase in the degree of freedom of expression due to changes in attached words and insertion of phrases. Furthermore, since word and phrase candidates are extracted by referring to the knowledge of the dictionary and the syntax / semantic information, when the input of a certain word is confirmed, the dictionary information about that word can be acquired, and as a result, morphological analysis and partial Parsing / semantic analysis is not required, and the parsing speed can be improved.

[0008]

Embodiments Embodiments in which the present invention is applied to an input unit of a database search system in Japanese will be described. The input of the system described in the present embodiment is a Japanese sentence mainly instructing a database search, and is generally a Japanese sentence including a search target, a search condition, and an edit instruction of a search result.

The operation instruction menu that the user can select is
As shown in FIG. 7, there are four types of word input instructions, word confirmation instructions, word deletion instructions, and search execution instructions.

The word input instruction is for estimating a word candidate and displaying the estimation result. The word confirmation instruction is for confirming the input word. The word deletion instruction is for deleting the input word. The search execution instruction is for analyzing the input sentence, executing a database search based on the result, and displaying the search result.

FIG. 1 is a diagram showing an outline of processing of this embodiment. In FIG. 1, selection of a word input instruction (step 1),
Although the selection of the word confirmation instruction (step 4) and the selection of the word deletion instruction (step 5) are sequentially described, any one of them may be actually selected depending on the situation.

When inputting a Japanese sentence, the user first selects a word input instruction (step 1). Then, in the word candidate estimation 2, the word dictionary 1 storing the word information
1, a syntax / semantic definition table 12 that defines constraint information about syntax / semantics that can be accepted as an input sentence, and a word table 13 that stores dictionary information of words input by a user.
With reference to, the syntactically and semantically correct word candidates are estimated (step 2). The syntax / semantic definition table 12 will be described in detail later.

Next, the words selected as candidates as the estimation result are displayed to the user (step 3). At this time, the first candidate is displayed in a manner different from the others. The user selects another candidate when the displayed first candidate is not the word intended by the user.

Next, a word confirmation instruction is selected to confirm the input of the word (step 4). At that time, in the syntax / semantic information storage 6, the dictionary information of the determined word and the information on the syntax / semantic definition table 12 are stored and held in the word table 13 (step 6).

In addition, when deleting a word that has been confirmed or selected due to a selection error or the like, when the word deletion instruction is selected (step 5), one word at the end is deleted. If the deleted word is a confirmed word, the word information stored in the word table 13 is deleted.

By appropriately repeating the processing from step 1 to step 6, words are sequentially input and confirmed.

When the input of the Japanese sentence is completed, that is, when the search execution instruction is selected, the syntax / semantic definition table 12 and the word table 13 are referenced to check whether the input sentence is correct ( In step 8), if there is no contradiction (step 9), the input is ended, and if there is an error, an error message indicating that is output (step 10) and the input is prompted again.

As is apparent from FIG. 1, the word candidate estimation 2, the word candidate output 3, the syntax / semantic information storage 6, the input sentence check 8, and the error message output 10 represent processing, and the word dictionary 11, the syntax / semantic definition. The table 12 and the word table 13 indicate files (also called tables).

As described above, according to this embodiment, each functional block is configured by the program logic. Therefore, an LSI can be provided for each functional block unit, and the processing speed can be increased as a word / phrase input device.

FIG. 2 is a block diagram of the hardware of this embodiment.

Reference numeral 31 is a selection device for selecting a word candidate or an operation instruction menu for inputting, deleting, confirming a word, and the like. In the present embodiment, these are selected by the mouse, but as another method,
It is also possible to use a touch panel that is selected by directly touching a finger on the screen, a method of using scroll keys (arrow keys) or number keys of the keyboard, or the like.

A display device 32 is for displaying word candidates and the like to the user.

Reference numeral 40 denotes a processing device for performing arithmetic processing based on a program stored in the storage device 50 while exchanging data with the storage device 50, the selection device 31, and the display device 32.

Reference numeral 50 denotes a storage device, which comprises the following storage areas. Reference numeral 51 is a word dictionary storage area, 52 is a syntax / semantic definition table storage area, 53 is a word table storage area, 54 is a word candidate estimation processing unit storage area, 55 is a syntax / semantic information storage processing unit storage area, and 56 is a working area. Is represented.

FIG. 3 is a diagram showing the contents of the syntax / semantic definition table 12 used in this embodiment. The syntax / semantic definition table 12 defines constraint information about the syntax acceptable to the system and the meaning between words. In this embodiment, as one description method of these constraint information, FIG.
The state transition diagram shown in is adopted. Of course, as another description format, an extended transition network (ATN) or the like can be easily substituted.

The state transition diagram of FIG. 3 consists of nodes and arrows. Each node has a unique name (N in FIG. 3).
1, N2, etc.) and each arrow can have an attribute. In this embodiment, the part-of-speech label representing the part-of-speech information is defined as an attribute.

The system uses the
It holds a pointer that transits the state transition diagram. The system transitions the pointer from node to node in the state transition diagram in response to a word input confirmation instruction or deletion instruction from the user.

Immediately before the user first inputs a word, the pointer is located at the head node N1 of the state transition diagram. If the pointer is at the node N3 at the end of the state transition diagram when the user finishes inputting the word, it is determined that the input was made correctly, and the sentence is treated as an acceptable sentence.

At a certain point of time, the user can input only the word having the part-of-speech label of the arrow having the node pointed to by the pointer as the starting point. The part-of-speech label is composed of terminal symbols such as nouns and verbs that cannot be further expanded, and non-terminal symbols such as [noun phrase] that can be expanded in more detail. When the part-of-speech label is a non-terminal symbol, it jumps to another state transition diagram that defines the method of expanding the part-of-speech label, and transitions in the transition diagram of the jump destination until it hits the terminal symbol. When the end node is reached, the state transition diagram immediately before the jump is returned to.

In this embodiment, the pointer transits only when the word is fixed or when the word is deleted. When a word is fixed, the pointer is moved from node to node in the arrow direction based on the part of speech of the fixed word. In addition, when a word is deleted, it moves backward to the node corresponding to the previous word in the direction opposite to the arrow. If such a transition is repeated and the pointer reaches the final node (node N3 in FIG. 3) when the user finishes inputting, the sentence is accepted as a word that the system can analyze. The process for checking whether or not the pointer has reached the final node is the input sentence check 8 in FIG. 1. If the final node does not have the pointer, the user is notified with a message indicating that the input sentence is not accepted. , Prompt again for a Japanese sentence that the system can accept.

As described above, by defining the syntax / semantic information of the system-acceptable sentence in advance and restricting the word input by referring to this information, only the system-acceptable sentence can be acquired as the input sentence. Therefore, the efficiency of analysis is improved. Further, by holding the dictionary information and the syntax / semantic information at the time of word input, it is not necessary to perform morphological analysis or part of the syntax / semantic analysis again, and the processing speed is improved.

FIG. 4 is a diagram showing the word dictionary 11 used in this embodiment. In the word dictionary 11, word headings 11
01 and part of speech 1102 are described. Further, item classification 1103 and corresponding column name 1104 are described as information necessary for database search.

FIG. 5 shows a word table 1 used in this embodiment.
It is the figure which showed the structure of 3. The word table 13 is a table that stores word dictionary information relating to a fixed input word and syntax / semantic definition table information corresponding to the word. That is, as the word dictionary information, the headline 130
2, part of speech 1303, item classification 1304, corresponding column name 1
The corresponding transition destination node 1306 is stored with 305 as the corresponding syntax / meaning definition table information. Here, the corresponding transition destination node 1306 is the node name (number) of the transition destination when the word is fixed and the pointer transits between nodes.
Represents

FIG. 6 is a diagram showing an example of an input Japanese sentence that satisfies the constraints of the state transition diagram described in FIG. In the example sentence 21 of (1), the pointer jumps from the node N1 to the node N4 by the noun "refrigerator", and further, the node N
Transition to 5. Next, the particle "no" causes a transition to the node N6, where the state transition for the noun phrase ends.
Further, the node N6 jumps back to the node N2 and returns. Similarly, for the noun "price" and the particle "wo", N2
→ N4 → N5 → N6 → N2 transition between nodes in this order.
Next, by the verb “search”, the node N2 transits to the node N3. Input ends here, but since the pointer is at N3 at this time, this sentence is regarded as a correct sentence,
Accept. If the example sentence 21 is “refrigerator price”, the pointer moves only to the node N2.
Such a sentence is not accepted and a message indicating that fact is output.

The example sentences (2) and (3) in FIG. 6 can be similarly accepted. In the following description, the example sentence 21 of (1) will be used.

FIG. 7 is a diagram showing a display screen in this embodiment. The display screen 101 has a large portion 102 for displaying a search instruction sentence, a portion 103 for displaying word candidates and search results, and a portion 10 for displaying a system message.
4, it is divided into a portion 105 for displaying an input operation menu. The input operation menu includes a word input 106 for instructing a word input, a word confirmation 107 for confirming the word input, a word deletion 108 for deleting the confirmed word or a word displayed as a candidate, and a search execution 109 for executing a search. Consists of types. Reference numeral 110 is a cursor, which is displayed below the word that is the current input operation target.

FIG. 8 is a PAD diagram (Problem Analysis Diagram) of the word candidate estimation process 2. Hereinafter, this process will be described according to the PAD diagram.

Whether or not the first word of the sentence is input is determined using a flag or the like (step 201). When the first word is input, a pointer to the state transition diagram of the syntax / semantic definition table 12 is determined. To the first node (node N1 in FIG. 3) of the transition diagram. Next, the word table 13 is initialized (step 203). After step 201, the word candidate extraction process (step 204) is called and executed. In the word candidate extraction process (step 204), first, the following process is performed for all arrows starting from the node pointed by the pointer (step 205). First, it is determined whether the part-of-speech label of the arrow is expandable (non-terminal symbol) (step 206), and if it is expandable, the pointer is jumped to the head node of the state transition diagram representing the part-of-speech label (step). 207). Next, the word candidate extraction process is recursively called (step 204). Next, the pointer is returned to the node immediately before the jump in step 207 (step 208). If it is not expandable in step 206, the word dictionary 11 is referred to extract words having the same part of speech as the part-of-speech label, collect them as word candidates, and hold them in the working area 56 (step 209). After step 204, the word candidates are displayed on the screen (step 21
0), of the word candidates, the first candidate is displayed on the search instruction sentence display portion 102 (step 211).

FIG. 9 shows the PAD of the syntax / semantic information storage processing 6.
It is a figure, and is the process which updates the word table 13 according to a user's instruction of word confirmation or word deletion
Hereinafter, this processing will be described according to the PAD diagram.

It is determined whether the instruction from the user is the deletion of a word or the confirmation of a word (step 601). In the case of the word confirmation instruction, the word dictionary 11 is referred to for the relevant word information, and the syntax / semantic definition. The position information of the pointer to the state transition diagram is stored in the word table 13 by referring to the table 12 (step 602). Next, the pointer to the state transition diagram is advanced by one based on the part-of-speech information of the decided word (step 603). Next, the displayed cursor is moved to the position immediately after the confirmed word (step 604). By deleting the cursor under the word, the user is notified that the input of the word is confirmed. Another method of notifying the user whether or not the word has been fixed can be easily replaced by a method of separating the display color of the unfixed word from the display color of the fixed word.

If the instruction from the user is a word deletion instruction in step 601, it is determined whether or not the last word currently displayed as the search instruction sentence is a fixed word (step 605). If it is a confirmed word, first, the information in the word table 13 corresponding to the word is erased (step 606). Next, referring to the word table 13, the pointer to the state transition diagram is returned to the corresponding transition destination node 1306 for the word immediately preceding the word (step 60).
7). After step 605, the word displayed in the search instruction text display portion is erased (step 608), and
The cursor is moved to immediately after the last word displayed (step 609).

A specific input method will be described below. Here, the operation method performed when the user inputs the example sentence 21 of (1) of FIG. 6 will be described below. Example sentence 2
Input can be realized in the same manner for 2 and 23.

First, the word input 105 is selected with the mouse (step 1 in FIG. 1). Then, the word candidate estimation 2 is activated. Initially, the pointer is at N1, but the adjective 86, the adjunct 87, and the noun 88 are extracted as the part of speech of the candidate word by the search in step 205 and thereafter. Step 20
9, the word dictionary 11 is referred to, and the word headings 1101 having these parts of speech are extracted and held as candidates.
Then, these word candidates are displayed as shown in FIG. 10, and the first candidate among the candidates (in the present embodiment, the dictionary 1 in FIG.
“Washing machine” (according to the description order of 1) is displayed in the search instruction text display portion 102 as shown in FIG. At this time, the cursor 110 is placed under the word "washing machine" that is the current input operation target.

Since the user wants to input "refrigerator" instead of "washing machine", another candidate "refrigerator" is selected with the mouse here. Then, the display is as shown in FIG. When "refrigerator" is selected, the display of the search instruction sentence display portion 102 also changes from "washing machine" to "refrigerator".

Since the user was able to change to "refrigerator", the user selects the word confirmation 107 with the mouse to confirm the input of "refrigerator" (step 4 in FIG. 1). Then, the syntax / semantic information storage 6 is activated. Step 602
By the word information about "refrigerator" and "refrigerator"
The node information corresponding to is stored as shown in FIG. Then, as shown in FIG. 12, the user is notified that the input of “refrigerator” has been confirmed by moving the cursor under “refrigerator” immediately after that.

Since the input of "refrigerator" is confirmed up to this point, the word input is again selected with the mouse in order to input the next word "no". Here, in order to reduce the number of menu selections, the word “refrigerator” may be automatically fixed when the word input is selected in a state where the word is not fixed (the state of FIG. 11). Alternatively, the next word candidate may be estimated and displayed at the same time when the word confirmation is selected.

Currently, the pointer to the state transition diagram of FIG.
Located at node N5. The arrow starting from the node N5 is the arrow 89, and the part-of-speech label of the arrow represents a particle.
Therefore, by using the word candidate estimation 2, the particle candidate is converted into the word dictionary 1
It is extracted from 1 and the word candidate including the first candidate "ga" is output. Since the user wants to input "no", he selects "no" from other candidates with the mouse. Then, it becomes as shown in FIG. When the input of "no" is confirmed, the pointer transits to the node N6, but since the transition diagram of the noun phrase ends here, the pointer returns to the node N2 next to the node N1 immediately before the jump.

The next words "price" and "o" can be input in the same manner (FIG. 14). At the time when “” is confirmed, the pointer is again in the node N2. The last word "search"
Can be similarly input, and the pointer finally transits to the node N3.

Since the input is completed here, the search execution instruction 109 is selected next.

Next, the input sentence is checked (step 8). When the input of the example sentence 21 is completed, the pointer is located at the node N3, which is the final arrival point, so the example sentence 21 is regarded as correct as the input sentence.

After that, the input sentence is analyzed, the database search is executed (FIG. 15), and the search result is displayed on the screen 103.

As described above, according to the present embodiment, the input sentence intended by the user can be created without key input, and even a person who is not good at keyboard type can efficiently input. In addition, words can be input and deleted word by word. Further, every time the input word is fixed, the word information of the word is extracted from the word dictionary 11 and stored and held. Therefore, it is not necessary to perform a morphological analysis or a part of the syntax / semantic analysis after the input is completed. Speed is improved.

The word input according to this embodiment is particularly effective for an information processing system having a limited number of vocabulary words and limited tasks. Even if the number of vocabularies used is relatively large, task-dependent semantic information (for example, in a database search, information that represents a word that corresponds to a column name in the database and a word that corresponds to a column value that belongs to the column as a pair)
If you further limit the word candidates using, or classify and display the system functions when displaying word candidates, or sort and display in alphabetical order or alphabetical order, it will be quite effective. Can be expected.

Modifications / extensions of this embodiment will be described below.

(1) When a word input instruction is selected while a word candidate is displayed (not confirmed), the word candidate is automatically confirmed, and the next word candidate is estimated and displayed. To do. As a result, it is only necessary for the operation menu to have two word input instructions and word deletion instructions,
Operation becomes easy.

(2) When inputting a Japanese sentence, an input example of the Japanese sentence can be provided as a catalog. If you do not know what input text should be input, you can input Japanese efficiently by referring to the catalog. Further, by making it possible to use the sentences in the catalog as they are or to use a part of them, the efficiency of inputting Japanese sentences can be further improved.

(3) In order to support a system having a large number of vocabularies, some words and phrases can be input with a keyboard or the like. With this, when there are many word candidates, keyboard input is possible, and the input efficiency is improved by using the input together with the selection of candidates. Also, in addition to keying in the phrase as it is, its synonyms, or,
By inputting a part of the phrase, candidates for the input phrase can be narrowed down to a small number.

(4) The present invention can be applied to a guidance system such as a tourist guidance system, an electric home appliance such as a video recorder, and the like, to an information processing system in which operation instructions are relatively limited and the vocabulary is small.

[0059]

According to the present invention, a syntactically and semantically correct input sentence can be generated without inputting a character from the keyboard. Moreover, it is possible to flexibly deal with an increase in the degree of freedom of expression due to changes in attached words and insertion of phrases. Furthermore, since the word information can be acquired at the time of input, it is not necessary to perform the morphological analysis and part of the syntax / semantic analysis again, and the natural language analysis processing speed is improved.

[Brief description of drawings]

FIG. 1 is a flowchart showing an outline of processing.

FIG. 2 is a block diagram showing a hardware configuration.

FIG. 3 is an explanatory diagram showing the contents of a syntax / sense definition table.

FIG. 4 is an explanatory diagram showing a word dictionary.

FIG. 5 is an explanatory diagram showing the structure of a word table.

FIG. 6 is an explanatory diagram showing an input Japanese sentence.

FIG. 7 is an explanatory diagram showing an input operation unit and a screen display unit.

FIG. 8 is a PAD diagram of word candidate estimation processing.

FIG. 9 is a PAD diagram of a syntax / semantic information storage process.

FIG. 10 is an explanatory diagram of display of word candidates.

FIG. 11 is an explanatory diagram of a display when another candidate is selected.

FIG. 12 is an explanatory diagram of a display of a confirmed word.

FIG. 13 is an explanatory diagram of a display when another candidate is selected.

FIG. 14 is an explanatory diagram of a display when another candidate is selected.

FIG. 15 is an explanatory diagram of a display during execution of search.

[Explanation of symbols]

2 ... Word candidate estimation processing, 6 ... Syntax / semantic information storage processing,
11 ... Word dictionary, 12 ... Syntax / semantic definition table defining syntax / semantic information, 13 ... Word table.

Claims (11)

[Claims] 1. A dictionary, an output means for displaying a candidate of a phrase to be input, a selecting means for selecting one of the displayed candidate words, a confirming means for confirming the input of the phrase, and the selecting means. In an input device provided with a deletion unit for deleting an input phrase, a unit for storing knowledge defining a constraint on a syntax of an input sentence or a constraint on a meaning of a phrase forming the input sentence, and the dictionary and the knowledge are referred to. By doing so, a phrase inputting device is provided, and means for estimating a phrase that can be input and displaying the phrase extracted by the estimation as a candidate. 2. The word / phrase input device according to claim 1, wherein the knowledge includes information about a part of speech of a word connectable to a word having a specific part of speech. 3. A means for referring to the syntax / semantic information defined by the knowledge, each time a word or phrase is fixed by the fixing means, or whenever a word or phrase is deleted by the deleting means. A word / phrase input device having means for updating and storing dictionary information and syntax / semantic information regarding a word / phrase fixed up to the point of time. 4. The phrase according to claim 1, further comprising means for notifying a message indicating that the input sentence is unacceptable when the input sentence that has been input conflicts with the syntax / semantic information defined by the knowledge. Input device. 5. The word / phrase input device according to claim 1, further comprising means for classifying and displaying the word / phrase candidates according to parts of speech. 6. The word / phrase input device according to claim 1, further comprising means for rearranging and displaying the word / phrase candidates in alphabetical order or alphabetical order. 7. The word / phrase input device according to claim 1, further comprising means for displaying the word / phrase selected by the selecting means at the beginning of word / phrase candidates from the next time onward. 8. The word / phrase input device according to claim 1, further comprising an example sentence collection in which an input example of a natural language sentence is stored in advance, and a means for referring to the example sentence stored in the example sentence collection when the natural language sentence is input. 9. The word / phrase input device according to claim 8, further comprising means for diverting all or a part of the example sentences stored in the example sentence collection as an input sentence. 10. The term input device according to claim 1, further comprising means for additionally providing means for key-inputting the term itself to be inputted, a part of the term to be inputted, or a synonym of the term to be inputted. 11. A dictionary, an output means for displaying a candidate word to be input, a selecting means for selecting one of the displayed word candidates, a confirming means for confirming the input of the word, and an input word In an input method in an input device having a deleting unit for deleting, by storing knowledge defining constraints on the syntax of an input sentence and constraints on the meaning of words and phrases forming the input sentence, by referring to the dictionary and the knowledge A method for inputting a phrase, comprising estimating a phrase that can be input and displaying the phrase extracted by the estimation as a candidate.