JP2002366543A - Document generation system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a document generation system to enable generation of a document without taking much trouble and time in correction of erroneous recognition by voice input even without using a keyboard. SOLUTION: Important words are extracted from a sentence inputted by voice and an erroneously recognized important word is estimated from the relation between the important words. A standard sentence most similar to the inputted sentence is selected among previously prepared standard sentences by using the remaining important words except the estimated erroneously recognized important word and used by changing part of it, if necessary. When a plurality of standard sentences are selected, a user selects and uses the most proper sentence.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a document creation system.

[0002]

2. Description of the Related Art A conventional document creation system will be described below as an example.

In recent years, a function for creating a document on a portable device or a portable telephone has been frequently used. Many devices are designed to create documents using small keyboards and buttons, but the keyboard is too small or only a few buttons need to be entered, which creates a problem in that it takes a lot of time and effort to create documents. ing.

In order to alleviate this problem, there has been proposed a function of retrieving a prepared sentence using a button or a cross key, and creating a document without a keyboard (Japanese Patent Application Laid-Open No. Hei 6 (1999)).
-216936). In this method, a sentence expression that is considered to be frequently used is prepared in advance, and the user selects a sentence expression to be used. For example, sentence expressions such as “1. Please call XX” and “2. Phone number is XX” are registered, and when creating a document informing the telephone number, the sentence expression of “2.” is registered. Select Next, a document is created by adding or modifying a part of the selected sentence. For example,
Complete the sentence by entering only the actual phone number, such as "Phone number is 012-3456". This function has reduced the time and effort required to create a document in a mobile device or mobile phone that is composed of input means including only buttons and cross keys.

[0005]

However, when there are many fixed phrases frequently used, it takes time to select the fixed phrases themselves. When there are many prepared sentences, it is usually divided into genres such as "greeting sentences" and "business sentences" to make it easier to select the fixed sentences. Is not uniquely determined, it is often necessary to select a sentence expression across a plurality of genres.

In addition, when the number of fixed phrases is limited so that a fixed phrase can be easily and quickly searched, a desired fixed phrase is often not prepared.

On the other hand, a method of creating a document by voice input has been proposed and put into practical use for a personal computer in order to reduce the trouble of inputting with a keyboard, but it takes time and effort to correct an editor when a recognition error occurs. There is a problem that it is not practical to use this in a mobile terminal that takes time.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above problems and to search for a sentence having the same or similar meaning as a sentence to be created by using an important word included in a sentence to be created. It is an object of the present invention to provide a document creation system that can output a sentence in which a part of a searched sentence is changed, thereby making it possible to create a document without using a keyboard and without taking time and effort. In addition, by using co-occurrence relationships and dependencies between key words to estimate the word that caused a recognition error, and removing the recognition error part and then performing sentence search, it takes time and effort to correct erroneous recognition by voice input. The present invention provides a document creation system that enables a document to be created without using a large amount of data.

[0009]

Means for Solving the Problems In order to solve the above-mentioned problems, a document creation system as a first means uses an important word extracting section for extracting an important word from a sentence, and using the extracted important words. And a similar example selection unit for selecting a similar sentence.

[0010] In order to solve the above-mentioned problems, a document creation system as a second means includes a voice recognition unit for converting an uttered sentence into a text, and an important word extraction for extracting an important word from the converted text. And a similar example selection unit for selecting a similar sentence using the extracted important words.

[0011] In order to solve the above-mentioned problem, a document creation system as a third means includes a voice recognition unit for converting an uttered sentence into a text, and an important word extraction for extracting an important word from the textized sentence. Part, a misrecognized word estimator that estimates misrecognized words from the extracted important words, and a similarity example that selects a sentence that expresses similar sentence using only important words other than the estimated important words And a selection unit.

[0012] In order to solve the above-mentioned problems, a document creation system as a fourth means includes a voice recognition unit for converting an uttered sentence into a text, and an important word extraction for extracting an important word from the textized sentence. And a misrecognized word estimation unit that estimates misrecognized words using the dependencies or co-occurrence relationships of the extracted important words, and a similar sentence using only important words other than the estimated important words. And a similar example selecting unit for selecting a sentence to be represented.

In order to solve the above-mentioned problems, a document creation system as a fifth and a sixth means includes a display for displaying a similar sentence selection result, and a display for confirming and correcting the selected similar sentence or a plurality of similar sentence selections. It is composed of buttons, numeric keys, or cross keys for selecting an optimal similar sentence when a sentence is selected.

[0014] In order to solve the above-mentioned problem, a document creation system as a seventh means comprises a speech recognition unit for converting a spoken sentence into a text, and a singular or plural similarities selected from the text-formed sentence. A display for displaying a sentence, and a selected similar sentence for confirming and correcting the selected similar sentence or selecting an optimal similar sentence when a plurality of similar sentences are selected.
It consists of buttons, numeric keys, or cross keys.

In order to solve the above-described problem, a document creation system as an eighth means includes an important word extracting unit for extracting an important word from an input sentence, and an important word for storing an already input important word. Utterance buffer, a next utterance prediction unit for predicting the field of a sentence to be input next using the key words held in the key word buffer, and a field selection unit for selecting a similar sentence suitable for the predicted field. Is done.

Next, the operation of each means described above will be described.

In a document creation system as a first means, a sentence having the same or similar meaning as the sentence to be created is searched for using an important word included in the sentence to be created.
By outputting the searched sentence or a sentence in which a part of the searched sentence has been changed, it is possible to create the sentence you want to create by inputting only the important words. And a document can be created.

A document creation system as a second means is as follows.
In addition to the operation of the document creation system as the first means,
Furthermore, by converting the spoken sentence into text by speech recognition and searching for a similar sentence from the text-sentence, even if a part other than the important word causes a recognition error,
Since a desired document can be created, the desired document can be created without taking the trouble and time required to correct erroneous recognition.

A document creation system as a third means is as follows.
In addition to the operation of the document creation system as a second means,
Furthermore, from the key words included in the speech-recognized sentences,
By estimating important words that have been misrecognized and searching for sentences that have similar meanings using only the important words other than the estimated important words, the parts other than the important words and some important words are mistaken. Even if recognition occurs, a desired document can be created without the trouble and time required to correct the misrecognition.

A document creation system as a fourth means is as follows.
In the misrecognized word estimation operation as the third means, the relationship between other important words included in the same sentence and the left By estimating an unrecognized important word as a misrecognized word, even if a part other than the important word and some important words cause misrecognition, it can be performed without taking the time and effort to correct the misrecognition further. Documents can be created.

The document creation system as the fifth means is as follows.
In the first operation, the similar sentence selection result is displayed on the display, and the similar sentence selected with the button, the numeric keypad, or the cross key is checked and corrected, and the sentence to be created by inputting only the important words is displayed. , It is possible to create a document without taking much time and effort.

A document creation system as a sixth means is as follows.
In the operation as the first means, when a plurality of similar sentences are searched, a sentence that the user wants to create by inputting only the important words by selecting an optimal candidate with a button, a numeric key, or a cross key. , It is possible to create a document without taking much time and effort.

A document creation system as a seventh means is as follows.
In the operation as the second means, by inputting the sentence to be created by voice input, displaying the similar sentence selection result on the display, confirming and correcting the selected similar sentence with the button, numeric key or cross key, it is important Even if a recognition error occurs in a part other than a word, a desired document can be created. Therefore, a desired document can be created without additional trouble and time for correcting the erroneous recognition.

The document creation system as the eighth means is as follows.
In the operation as the first means, the input key words included in the already processed utterance are used to predict the field of the next input sentence, and priority is given to similar sentences suitable for the predicted field. By making the presentation, it is possible to input with less errors such as recognition errors, and moreover, it is possible to present the optimum sentence accurately, thereby making it possible to create a desired document without taking time and effort. .

[0025]

Embodiments of the present invention will be described below with reference to the drawings.

(Embodiment 1) FIG. 1 is a block diagram of a document creation system according to an embodiment of the present invention, and relates to the above-described first to seventh means. FIG.
Is a document creation system diagram.

First, before creating a document, the dependency structure of the corpus is analyzed in advance, and key word pairs having a dependency are extracted.
It is described in the Keyword Dependency Table. The analysis of the dependency structure may be performed using an analysis tool or may be performed manually. For example, from the corpus shown in FIG. 4A, a relation table as shown in FIG. 4B is described. In this case, instead of the important words themselves, those having the same concept as the important words may be classified into the same class, and described in the table as dependencies between the classes. For example, the dependency relationship between “tomorrow” and “holding” in FIG. 4B becomes clear from the classification vocabulary table that “tomorrow” is a vocabulary included in the concept of “date”. As shown in FIG. 4C, the information may be described in a dependency relationship of “date class” and “holding”. By describing in the concept class in this way, not only “tomorrow” but also other vocabulary (today, tomorrow, etc.) included in “date” can be handled, and thus there is an advantage that a wide range of input can be handled.

Next, for the sentences included in the corpus, the correspondence between the important words included in each sentence and the sentences is clarified and described in the example sentence DB. FIG. 4D shows an example of the example sentence DB.

Next, when creating a document, the voice input key 2
Press 2 and the voice of the sentence you want to create
, And the contents are converted into text by the speech recognition unit 1. Next, an important word is extracted from the text converted by the important word extracting unit 2. The important words at this time may be selected in advance from the dictionary, or content words such as nouns and verbs may be used as important words. Next, the misrecognition estimating unit 3 compares the key word dependency relation table 6 extracted in advance with the key words extracted from the input text, and determines the key word pairs described in the dependency relation table in the input text. Check if it exists. As a result, when the important word present in the input text does not have a dependency with any other important word existing in the same input text, the important word is estimated to be a misrecognized word.
For example, in the dependency table, the key word pairs (apple, eat) and (seat, sit) are described as dependencies.
When the text "Eat an apple at a seat" is output as a recognition result, the important words "seat" and "apple" are output.
When comparing “eating” and the dependency table, no dependency is found for “apple” or “eat” for “seat”. At this time, it is estimated that “seat” is a misrecognized word.

Next, the example selecting unit 4 uses the remaining important words except for the important word that is presumed to be misrecognized, and sets the example sentence that waits for the most important keyword pair having the same dependency as the input sentence. The sentence is extracted as being regarded as an example sentence having the same intention, and the example sentence is output to the display 21. If a plurality of example sentences are selected, a plurality of example sentences are extracted. Next, the user checks the example sentence output on the display 21, and presses the cross key 23 when determining that the sentence can be accepted as a document to be created by the user. When a plurality of example sentences are output, the cursor is moved to the optimal example sentence with the cross key 23 and the key 23 is pressed. If an appropriate example sentence is not displayed, move the cursor to the closest example sentence, correct it with the numeric keys 25, and press the confirmation button. For example, a word may be added to the end of "Please love your body" and modified to "Please love your body." If no closest example sentence has been output, the voice input key 22 is pressed to perform voice input again. The document for which the key 23 is pressed is held as a confirmed document, and the user repeats inputting the next document.

If a desired document cannot be selected even if the voice is input many times, the document may be input using the numeric keys as in the related art.

In the above embodiment, a sentence to be created is input by voice, and a sentence expressing the same or similar sentence is selected from the example DB using the important words included in the input sentence, and the selected sentence or the selected sentence is selected. By outputting a sentence in which only a part of the sentence is changed, it is possible to create a document on a portable device or the like that does not use a keyboard without taking much time and effort.

(Embodiment 2) Next, a document creation system according to another embodiment will be described. FIG. 3 is a block diagram of a document creation system according to an embodiment of the eighth means.

First, before creating a document, an important word dependency relationship table and an example sentence DB are created as in the previous embodiment.
At this time, the original corpus is classified according to the field, and field information of the corpus including the sentence used for the extraction is described in each of the extracted dependencies and the example sentence DB.
For example, field information such as "greeting" is added to the DB extracted from the greeting document corpus, and field information such as "notification" is added to the DB extracted from the notice document to the dependency relationship table and the example sentence DB (FIG. 5A). 5 (b)).

Next, when creating a document, the first document is created by performing the same processing as in the previous embodiment. However, the field of the example sentence DB finally selected by the user is stored in the field history table. When inputting a second document and selecting an example sentence, the field prediction unit inputs information on the previous field from the history table, and when selecting an example sentence, preferentially selects an example sentence belonging to the corresponding field. I do. For example, if the example sentence DB selected at the time of the first document input belongs to the “information” field, the important words included in the second sentence input are:
First, compare it with the key word pair in the "Notice" field,
Example sentences are displayed on the display along with the field name. If the desired example exists, move the cursor and press the cross key as in the first utterance. However, if there is no key word pair corresponding to the comparison, move the cursor to the field name and When the key is pressed, the result of the selection is displayed again from the dependency table across fields. Similarly, when selecting an example sentence DB, first, the example sentence DB in the same field is searched, and if an appropriate DB is not selected, an example sentence DB across fields is searched. After the user selects the best example sentence,
As in the previous embodiment, the document is fixed by partially modifying it as necessary.

In the above embodiment, the search was performed by narrowing down the field in comparison with the important word pair and selecting the example sentence DB. However, the field may be narrowed down only in the case of selecting the example sentence DB. Further, in the above embodiment, first, the fields are narrowly selected, and the selection across fields is performed only when an appropriate important word pair or example is not selected. The field may be narrowed down only when the number of similar example sentences exceeds a certain number.

[0037]

As described above in detail, the first document creation system according to the present invention uses a keyword included in a sentence to be created to generate a sentence representing the same or similar sentence as the sentence to be created. By searching and outputting a searched sentence or a sentence in which a part of the searched sentence has been changed, it becomes possible to create the sentence you want to create by inputting only the important words. It is possible to create a document without spending much time.

Further, the second document creation system according to the present invention, in addition to the operations described in the first document creation system,
Furthermore, by converting the spoken sentence into text by speech recognition and searching for a similar sentence from the text-sentence, even if a part other than the important word causes a recognition error,
Since a desired document can be created, the desired document can be created without taking the trouble and time required to correct erroneous recognition.

Further, in the third document creation system according to the present invention, in addition to the operation of the second document creation system, the third document creation system is erroneously recognized from the key words included in the speech-recognized sentence. By estimating important words and searching for sentences that have similar meanings using only the important words other than the estimated important words, even if parts other than the important words and some important words cause misrecognition, In addition, a desired document can be created without the trouble and time required to correct erroneous recognition.

Further, the fourth document creation system according to the present invention is characterized in that, in the misrecognized word estimation operation of the third document creation system, the dependency or the co-occurrence between important words included in the speech-recognized sentence. By presuming that important words in the same sentence that do not have the same relationship as the left are recognized as misrecognized words, the non-important words and some important words may be misrecognized. Even if it occurs, a desired document can be created without the trouble and time required to correct erroneous recognition.

In the fifth document creation system according to the present invention, in the operation of the first document creation system, a similar sentence selection result is displayed on a display, and a similar sentence selected by a button, a numeric key, or a cross key is displayed. By confirming and correcting, it is possible to create a sentence to be created by inputting only important words, so that a document can be created without taking much time and effort.

Further, in the sixth document creation system according to the present invention, in the operation of the first document creation system, when a plurality of similar sentences are searched, the user can use a button, a number key, or a cross key to select an optimum one. By selecting a candidate, it is possible to create a sentence to be created by inputting only important words, so that a document can be created without taking much time and effort.

Also, in the seventh document creation system according to the present invention, in the operation of the second document creation system, a sentence to be created by voice input is inputted, a similar sentence selection result is displayed on a display, and the selected sentence is selected. By confirming and correcting similar sentences with buttons, numeric keys, or cross keys, it is possible to create a desired document even if a part other than an important word has a recognition error. A desired document can be created without spending time.

Further, in the eighth document creation system according to the present invention, in the operation of the first document creation system, the next input word is input using the input important word included in the already processed utterance. By predicting the field of the sentence to be given and preferentially presenting similar sentences suitable for the predicted field, it is possible to input with less errors such as recognition errors, and it is possible to accurately present the optimal sentence As a result, a desired document can be created without taking time and effort.

[Brief description of the drawings]

FIG. 1 is a block diagram of a document creation system according to an embodiment of the present invention;

FIG. 2 is a diagram of a document creation system according to an embodiment of the present invention.

FIG. 3 is a block diagram of a document creation system according to another embodiment of the present invention;

FIG. 4 is a diagram for explaining analysis of a dependency structure.

FIG. 5 is a diagram for explaining analysis of a dependency structure.

[Explanation of symbols]

 1 Speech Recognition 2 Key Word Extraction 3 False Recognition Estimation 4,12 Example Sentence Search 5 Output Sentence Creation 6,10 Key Word Dependency Relationship Table 7,11 Example Sentence DB 8 Field Prediction 9 Field History Table

──────────────────────────────────────────────────の Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G10L 15/00 G10L 3/00 561C 15/18 537C 15/22 551B F-term (Reference) 5B009 KB00 ND04 VA02 5B075 ND03 PP07 PQ02 PR06 QM08 QM10 QP01 UU40 5D015 HH00 KK02 LL04 LL08

Claims (8)

[Claims] 1. A sentence expressing the same or similar sentence as a sentence to be created by using a minimum number of words (hereinafter, referred to as an “important word”) necessary for understanding the meaning included in the sentence to be created. A document creation system for searching and outputting a searched sentence or a sentence obtained by changing a part of the searched sentence. 2. The document creation system according to claim 1, wherein the uttered sentence is converted into a text by voice recognition, and a similar sentence is searched for the transcribed sentence. 3. A sentence representing a similar sentence by estimating an erroneously recognized important word from important words contained in a speech-recognized sentence and using only important words other than the estimated important word. 3. The document creation system according to claim 2, wherein 4. A method in which a dependency or a co-occurrence relationship between important words included in a sentence recognized in a sentence is used and the dependency or a co-occurrence relationship with another important word included in the same sentence is not recognized. The document creation system according to claim 3, wherein the word is estimated to be a misrecognized word. 5. The document creation system according to claim 1, wherein a similar sentence selection result is displayed on a display, and the similar sentence searched by a button, a numeric key, or a cross key is confirmed and corrected. 6. The document creation system according to claim 5, wherein, when a plurality of similar sentences are searched, a user selects an optimum candidate using a button, a numeric key, or a cross key. 7. A sentence to be created by voice input, a similar sentence selection result is displayed on a display, and a similar sentence searched by a button, a numeric key, or a cross key is confirmed and corrected. Document creation system described in. 8. The method according to claim 1, wherein a field of a sentence to be input next is predicted by using an already input important word, and a similar sentence suitable for the predicted field is preferentially presented.
Document creation system described in.