JPH0258115A - Interactive device - Google Patents

Abstract

PURPOSE:To detect a part which a system cannot analyze in a user input sentence at an early time point, and to attain smooth interaction by analyzing a character string in the middle of an input even when the input of a sentence is not completed, and responding at such a time point. CONSTITUTION:An input part 1 to input a natural language sentence by a user, an analyzing part 2 to determine the meaning of the input character string for the character string obtained from the input part 1, a problem solving part 3 to solve a problem based on an analysis result and to determine a response to the user, and a responding part 4 to output responding contents are provided. Consequently, before an input completion mark is inputted by the user, the already inputted character string can be analyzed. Thus, the part which an interactive device cannot analyze can be recognized at the early time point, and a processing according to the part can be executed.

Description

[Detailed Description of the Invention] [Purpose of the Invention] (Field of Application in M Business) This invention enables users to interact with a system using a natural language such as English or Japanese using a keyboard or tablet. , regarding dialogue rewards that promote problem solving.

(Prior art) In an interactive device that solves problems by using natural language dialogue with a computer, when the system responds to an input request by the user's turn, the system normally responds with a special message indicating the end of input. The contents of the user's input are not processed until the end of the input is declared by manually entering a symbol (generally referred to as a manual end symbol, such as ΦYear Ridge Return or Line Feed).

In this type of IT, the system cannot detect the unresolved EM and unrefined concepts that exist in the user's input sentence until the input is completed and the next analysis phase is started.
The parts entered by the user on the premise that the system will understand the unknown words and unsolved thoughts of Kotori are wasted or have to be re-entered, increasing the burden on the user and slowing down the dialogue. The smoothness of the system was impaired.

(Problem to be Solved by the Invention) As described above, in the conventional dialogue device t, even if there is a part in the input sentence that the system cannot analyze, it cannot be detected until the user finishes inputting. In such a case, there is a problem that the burden on the user is large and the dialogue itself cannot be smoothly carried out.

Therefore, an object of the present invention is to detect parts of a user's input sentence that cannot be analyzed by the system as early as possible, thereby enabling smoother dialogue.

[Structure of the invention]

(Means for solving the problem) This Seimei has an input section for the user to input natural language sentences, and performs morphological analysis, syntactic analysis, semantic analysis, etc. on the character string obtained from this input section. An analysis section that determines the meaning of an input character string, a problem-solving section that solves the ζζ problem based on the analysis result and determines a response to the user, and a response section that outputs the answer to the question from the problem-solving section. The character string input sequentially from the input section is analyzed by the analysis section before the user inputs the input end symbol, and depending on the result,
The feature is that it responds to the user even if the input is not completed.

(Operation) This invention determines whether the character input by the user in the input unit is a delimiter specified in advance by the system, and if so, the character inputted before this delimiter is sent to the analysis unit at that point. As a result of the analysis, kl)t@- was detected as an unknown word or unknown concept, or it was already sent to the analysis department and the results of the analysis department were enriched and a higher level analysis was performed. If it is determined that the input is inappropriate, the result is immediately sent to the problem analysis section and the user responds even before the input end instruction is given.

(Embodiment) First Embodiment FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, 1 is an input section, 2 is an analysis section, 3 is a problem analysis section, and 4 is a response section.

In this embodiment, a case will be described in which a character string input by the user, such as English or German (0), does not go through the conversion process 2 and becomes the character to be analyzed by the analysis unit 81. Figure 2 shows
FIG. 3, which is a block diagram of the input section, is a flowchart of the algorithm of the input control section.

The input unit first inputs one character from the character input means 11 and determines whether the character is a delimiter stored in the delimiter memory 1g14 (22). In English, spaces, commas, periods, etc. are used as examples, and input end symbols are also referred to as such. If it is not a delimiter, store that character in the character string-time storage unit 1323),
Return to 1 character input (21). If it is a delimiter, the character string that has been input so far is retrieved from the character string-time memory section and it is determined whether the content is empty (24). If it is not empty, the string is sent to the analysis section E (24). 25) Clear the contents of the storage section (26). Furthermore, it is determined whether the delimiter can be omitted, such as a space (27), and if it is not omissible, it is also sent to the analysis unit (28).

FIG. 4 is a configuration diagram of the analysis section, and FIG. 5 is a flowchart of the algorithm of the analysis control section.

In the analysis section, first input one word from the input section41),
The input muml is determined as the input end symbol (42).

If it is not an input end symbol, look up the word in a dictionary using Parsing Condolences 4I32, and if the dictionary lookup fails °
If so, immediately send it to the problem solving department (44). If the dictionary lookup is successful, it is determined whether the word is a sentence-final symbol such as a period or a question mark (
45) If it is not a sentence-final symbol, it is recorded in the word-time storage unit 34 and 2 (46). , if the input word is a sentence-final symbol, iic! Take out the recorded word string and use the grammar for analysis and 'f5$
33 to determine if it is acceptable as a sentence (51).

Therefore, I cleared the contents of the olfactory word-temporal memory (52
), and after that, if the parsing is successful, record its existence in the parsing result-time memory unit 35 (54), and if not, immediately notify the problem solving unit that the sentence parsing has failed (54). 55].

In the branch at step (42) above, if the input word is an input end symbol, first check whether the analysis result-time memory section is empty (60), and if it is not empty, check its existence. is sent to the problem solving section (61), and the analysis result-time storage section is cleared (62). Furthermore, it is checked whether the abbreviation-time storage section is empty (63), and if it is not empty, its contents are analyzed in the same manner as in step (51), and whether the analysis succeeds or fails, the result is The problem solving section ζ is sent (65).

The problem solving unit usually recognizes the user's instructions, requests, etc. based on the sentence analysis results sent from the analysis unit, and solves the problem accordingly. According to (55), even if the end-of-input symbol is not input, a message to that effect is immediately sent to the user who has failed in dictionary lookup or sentence analysis. It becomes possible to make requests and ask questions to the user.

Second Embodiment In the first embodiment, the character string input by the user does not go through the conversion process and becomes the character to be analyzed by the analysis unit as it is, such as in English or German. Next, we will show examples of cases where there is a strange odor, process, etc., such as in Japanese.

The overall layout is the same as in Figure 1. Figure 6;
FIG. 7, which is a block diagram of the input section, is a 70-chart of the algorithm for performing step (21) of the algorithm of the input control section shown in FIG.

As shown in FIG. 6, in this embodiment, a character string conversion means 71 is provided between the character input means in 8 of the configuration of the input section in the first embodiment and the human control section, and the conversion result thereof. is temporarily stored in the converted character string-time storage section 72.
has been added. Accordingly, the step of inputting one character (step 21) in the input algorithm (threshold 31) of the first embodiment is changed as shown in the flow shown in FIG.

First, when it comes to the step of inputting one character (step 21), it is checked whether the character string after conversion - the time memory is empty (
81), If it is not 2, take out the first character from Q] and use the character input in step 21.82)%. In step 81, if the - time storage section is empty, input each character one by one from the character input means 70, perform kana-kanji conversion, etc. (83), and store the result in temporary memory gCζ (84); The first character is extracted (82).

FIG. 8 is a flowchart of the algorithm of the analysis control section in the second embodiment.

In languages such as Japanese that do not have separate writing, the data sent from the input section must be analyzed as one or more clauses rather than words as in English. First, an input is made from the input section (91), and the human input end symbol is determined (92). If it is not an input end symbol, the delimiter is determined (93), and #a&, which is a character string that is not a delimiter, is analyzed to see if it can be recognized as one or more clauses (94). If the analysis is successful, record the analysis result in the single-time memory 96), otherwise'! Oki informs the problem solving department that the analysis has failed (97). Also, in step 93i, the input is found to be a delimiter '+1% f,
It is determined whether it is a sentence-final symbol (98), and if it is not, then, also in step 96, the number is memorized in the interrogative-hour memory section. If the input is a sentence-final symbol, it is analyzed whether the input from the connotation stored in the word-time memory unit to the sentence-final symbol can be accepted as a sentence.99) If the analysis is successful, it is used as the analysis result. If the sentence parsing fails, the sentence parsing failure is immediately notified to the multiple problem solving unit t.

In addition, in step 92, if the input is an input end symbol, as in the first embodiment, first, the content of the analysis result-time memory unit is sent to the problem solving unit. Analyze the contents of the section and send it to the problem solving section.

In this way, even in languages such as Japanese, where the characters that can be input using the character input means and the characters that can be analyzed are different, or when a sentence is written in a language that is not separated in advance, the end of input symbol is input. It is possible to perform the analysis before the interaction device starts, and to show the user at an early stage the parts that cannot be analyzed by the dialog device.

〔Effect of the invention〕

As explained above, according to the present invention, in the dialogue device It that solves problems through dialogue with the user, the character string that has already been input is analyzed before the input end symbol is manually input by the user. It is now possible to do
This makes it possible for the dialogue device to recognize parts that cannot be analyzed at an earlier stage and process them accordingly, greatly improving the efficiency of the dialogue.

[Brief explanation of the drawing]

1 is a schematic diagram showing one embodiment of the present invention, FIG. 42 is a block diagram showing an example of the structure of the input section in FIG. 1, and FIG.
The figure is a flowchart for explaining the operation of the input control section of FIG. 2, FIG. 4 is a block diagram showing an example of the configuration of the analysis section of FIG. 1, and FIG. 5 is the operation of the analysis control section of FIG. 4. FIG. 6 is a flowchart showing one configuration of the input section in the second embodiment of the present invention, and FIG.
The figure is a flowchart for explaining the operation of the character input part of the input control section shown in FIG. 6, and FIG. 8 is for explaining the operation of the analysis control section in the second embodiment. This is flowchart 1 for doing this. 1...Input section, 2...Analysis section, 3...Interval resolution section, 4.Response section ! Same as right Hikaru Matsuyama Figure 2 Figure

Claims (1)

[Claims] An input section for inputting a sentence, an analysis section for analyzing the character string of the sentence obtained from the input section, and a problem solving unit that solves a problem based on the analysis result obtained by the analysis section. It is equipped with a problem-solving part that determines the response content and a response part that outputs the response content of this problem-solving part, and even when the input of a sentence is not finished, it analyzes the string that is being input and uses the analysis result to An interaction device characterized by responding at a point in time.