JPH02255942A - Conversation type input/output method - Google Patents

Abstract

PURPOSE:To automatically learn the words which are not registered into a word dictionary by comparing the character strings to be inputted with those of words included in the word dictionary and selecting those words having common parts as the candidate words after displaying them. CONSTITUTION:A partial string matching part 2 compares the first one of input character strings with the head character strings of each word included in the word dictionary 11 and adds the common parts, if detected between both character strings, to a candidate list. This list is displayed at a candidate display part 3. A word register part 4 compares the input character string with the character string of a selected word and the meaning information and estimates a word to be inputted based on the input character strings so far inputted to register the estimated word into the dictionary 11. A word estimating part 5 compares an input word string with the word strings which are so far inputted and recorded in an input history 12. An unknown word estimating part 6 compares the words set before and after an unknown word if included in the so far input ones with the word strings recorded in the career 12 and estimates the meaning information on the unknown word.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an interactive input/output method for providing a flexible interface between a computer and a user.

[Prior Art] As a conventional input method to a computer, the input sentence from the user is generally processed as a string of words registered in advance in a word dictionary (for example, Japanese Patent Laid-Open No. 62 -22
1776). For this reason, all words that can be input by the user must be registered in advance in a word dictionary, which requires a dictionary with a huge number of words. Furthermore, if there is a possibility that proper nouns may be input, it is practically difficult to register all proper nouns in the word dictionary.
The input word will be treated as an unknown word, and the input from the user will remain uninterpreted.
Input sentences from users are interpreted according to the contents of a syntactic/semantic dictionary based on language models and world models. but,
It is not easy to develop a general and versatile dictionary that covers all kinds of expressions, and even if the task is limited, there are naturally limitations on the expression formats that users can input. The current situation is that

[Problems to be Solved by the Invention] In the above-mentioned conventional technology, the words input by the user and the method of expressing input sentences are limited to the word dictionary and language model prepared in advance in the system. There were problems such as not being able to input text or input sentences using a user-specific expression method.

An object of the present invention is to provide an interactive input/output method that allows a user to input information to a computer using a unique expression method.

[Means for solving the problem] In order to achieve the above purpose, a common name is recorded in a word dictionary along with semantic information indicating its category as a core for estimating proper nouns. It is.

Furthermore, in order to learn expression methods unique to the user, word strings input by the user are recorded as history information along with semantic information.

[The character string input by the action user is compared with the character string of each word in the word dictionary to determine whether there is a common part, and if there is a common part, the corresponding word is It is displayed as a candidate word for the user to select. At this time,
If the candidate word corresponds to a common name, the input character string can be estimated to be a proper noun corresponding to the common name, and learning can be performed, allowing the proper noun to be input naturally.

In addition, by comparing the word string input by the user with history information, which is the word string input so far, at the level of semantic information, the next input word string is adjusted according to the user's unique expression method. It becomes possible to predict and display the words that will be used in a given sentence, and to estimate unknown words contained in an input sentence, making it possible to support friendly input that matches the user's expression method.

[Example] Hereinafter, an example of the present invention will be described with reference to FIG. 1st
The figure is a system configuration diagram according to the present invention. Input from a keyboard or the like (not shown) is interpreted in the character input section 1. In the case of normal character input, control is passed to the substring matching unit 2, which checks whether there is a common part between the character string input so far and the character string of each word registered in the word dictionary 11. Compare.

FIG. 2 shows the processing flow of the subsequence matching unit 2.
First, the character string from the beginning (n=1) of the input character string is compared with the character string from the beginning of each word in the word dictionary 11, and if the character strings match up to the end of the input character string, Since there is a common part between the word and the input character string, the word is added to the candidate list. In this way, n of the input string
After comparing the character string starting from the th character string with the character string from the beginning of the word, the first step is to compare the character string starting from the (n+1)th character string of the input character string, and the same process is performed until the end of the input character string. Repeat. As a result, a set of words whose initial character string is common to the end of the input character string is obtained as a candidate list. For example, when the input character string is "Ounukai", it will include "r f's & N L company) J + ris bq do (road)" and "-ri-chiba (market)" (the underlined part is common parts).

The word candidate list obtained by the subsequence matching section 2 is displayed on the screen in a candidate display section 3 in the form of a pop-up window, a menu selection window, or the like. If the user selects one of the displayed word candidates by clicking the mouse,
Control is transferred from the character input section 1 to the word registration section 4.

The word registration unit 4 estimates the word to be input from the character strings input so far by comparing the character string and semantic information of the selected word with the input character string. In the above example, if "Kaido" is selected, it is estimated from the semantic information (recorded in the word dictionary) that "Kaido" is a general name representing a highway, and "Oume" is selected.
Learn the proper noun of the road named ``Ounukaido (Ome Road)'' and use the word dictionary 11.
Register. Furthermore, it is additionally registered in the input history 12 as the history of the user's input word string.

The word prediction unit 5 compares the input word string with the word strings that have been input so far recorded in the input history 12, and selects the next word string from the word strings whose semantic information matches. The category of the input word is predicted and displayed on the screen by the candidate display section 3. For example, we have already seen the ``Ome-type (
city)” and “What is the population? (city attributes)”, then click “
When "(city) in Kokubuzenfu" is input, a word representing the attributes of the city (population 2 area, etc.) is predicted to come next and displayed as a candidate for input word prediction.

When an unknown word is included in the input so far, the unknown word prediction unit compares the words before and after the unknown word with the word string recorded in the input history 12, and extracts the meaning information of the unknown word. presume. For example, similar to the example above, if the word strings "Ome-type (city)" and "What is the population? (city attributes)" are recorded in the input history 12, "Kunitachi-no (unknown word)" .

If there is an input such as ``What is the population?'', it can be estimated that it is ``R city called Kunitachi'', which is displayed as an unknown word candidate on the candidate display section 3, and unknown word resolution is performed.

Note that the partial sequence matching section 2. Word prediction unit 5゜Unknown word prediction unit 6. Conventional techniques such as linear matching can be used to compare character strings and word strings. However, by using non-linear matching, which takes into account omissions, insertions, and incorrect inputs of characters caused by people such as Kii people, as well as omissions of attached words such as case particles, it is possible to expand the scope of the system's estimation.

Below, a detailed example of operation in the system configuration shown in FIG. 1 will be described.

FIG. 3 shows the initial registered words and their meaning information in the word dictionary 11 (case particles are omitted here to simplify the operation).

Until the user inputs "ou", there is no common part between the input character string and each word in the substring matching unit 2, and the word candidate list is empty, so the candidate display unit 3 , it simply displays an echo back of the input characters. The input is "Oumu" (the underlined part indicates the newly input character.

), there is a common part with "menseki (area)" and it is displayed as a candidate word.

The input progresses further, and when the input is "Oume-ka-", "Kaija (company)" and "Kaido (road)" are displayed as candidate words.

At this point, if the user inputs an input to select "Kaido", it is clear that the user is trying to input "r Oume Kaido (Ome Kaido)", and furthermore, "Kaido" is a general road. Since it is a name, it is also clear that ``Ome Kaido'' is a proper noun that refers to a road. Therefore, in the word registration section 4, "Ome Kaido (Ome Kaido, proper noun, line)" is registered in the word dictionary 11 and the input history 12. As a result, the system has learned "Ome Kaido", and from now on, the input from the user will be "Ome Kaido".
0. ”, “Ome Kaido” is selected as one of the candidate words.

In this way, users can access [``Ome Kaido''].

Enter "Population", "Maximum", "Town"]
When performing a map information search, the string of characters that the user should input is ``Oumekahon to Honji*Samotomahon'' (this mark indicates candidate word selection input).

As a result, it is registered in the input history 12 as a word string corresponding to "line, line surface relation 2-surface attribute, attribute limited 2-surface".

Following this, the user selects [``Ome Street Floor'', ``Pass'', ``
If you try to search for "area", "minimum", "r town"],
In response to the input of "5", "Ome Kaido" is displayed as a candidate word, and when this is selected, the word prediction unit 5 selects "line", which is the semantic information corresponding to "r Ome Kaido", and the input history 12 A comparison is made with ``line'', and a word with the meaning information of ``r-line surface relation following ``line''(``through'' is registered in the word dictionary 11) is predicted as the next word input.
Displayed by the candidate display section 3. Furthermore, this “pass”
When the user selects , the word prediction unit 5 predicts ``1-page attribute'' through similar processing, displays ``population'' and ``area,'' and the user simply selects ``area.'' good. The same input/output control is performed below, so in order to search for "Ome Kaido,""through,""area, [minimum], r town"], the user simply needs to ”, that is, according to the display based on the system's estimates and predictions, according to the user's own expression method,
You can type naturally and efficiently.

Furthermore, if the user subsequently enters ``Fuchyuuka＊Yokogiruji＊'' with the intention of ``Fuchukaido'', ``Cross'', [Population],...'', Unknown word prediction unit 6
Now, [line] in input history 12.

From the comparison of the series ``Line surface relation 2-sided attribute, ......'' and the series ``Line, unknown word 2-sided attribute ko'' of the input sentence, ``
The unknown word ``Yokogiru'' indicates ``r-line surface relationship''.
It can be inferred that this word is similar in meaning to the word ``Toruru''. This is displayed on the candidate display section 3, and by the user's selection input indicating consent, the word "Yokogiru (crossing, verb, linear relation)" has been learned and is registered in the word dictionary 11. do.

[Effects of the Invention] According to the present invention, it is possible to automatically learn words that have not been registered in advance in a word dictionary, and it is also possible to predict input words according to the user's own phrasing. Therefore, input from the user to the computer can be performed naturally, efficiently, and friendly.

[Brief explanation of drawings]

FIG. 1 is a system configuration diagram of an embodiment of the present invention, FIG. 2 is a diagram showing the processing flow of a subsequence matching section, and FIG. 3 is a diagram showing the contents of a word dictionary. 1... Character input section, 2... Substring matching section, 3
... Candidate display section, 4... Word registration section, 5... Word prediction section. 6...Unknown word prediction unit, 11...Word dictionary, 12...
- Input history. Kaya 1 map

Claims (1)

[Claims] 1. In a conversation system in which input and output between a user and a computer are performed in a conversational manner, common names are also recorded in a word dictionary containing semantic information for interpreting input contents. , a conversation characterized by determining whether or not there is a common part between the input character string and the character string of each word in the word dictionary, and displaying the corresponding word as a candidate word if there is a common part. Type input/output method. 2. A claim characterized in that when a predetermined word is selected from among the displayed candidate words, the input character string is added to and recorded in a word dictionary and a history dictionary as words having the same meaning information. The interactive input/output method described in 1. 3. If a general name is selected from the displayed candidate words, the input character string is presumed to be a proper name of the type corresponding to the general name, and is added to the word dictionary and history dictionary. 2. The conversational input/output method according to claim 1, further comprising recording. 4. Compare the semantic information of the input word (or word string) with the semantic information of the word (or word string) in the provenance dictionary,
4. The conversational input/output method according to claim 3, wherein the next input word is predicted from matching words (or word strings) and displayed as candidate words. 5. Compare the semantic information of the input word string with the semantic information of the word string in the history dictionary, and estimate the semantic information of the unknown word included in the input word string from the matched word string. The interactive input/output method according to claim 3, characterized in that: