KR100276687B1 - Interactive Learning Aid and Its Conversation Analysis Method U - Google Patents

Abstract

The present invention relates to an interactive learning aid and a method of analyzing the same.

The present invention provides a pre-storage device for providing data for morphological and syntax analysis, meaning and discourse analysis, a knowledge base storage device for providing data for dialogue analysis, and a voice input through a microphone. A speech / text converter for converting text into text, and a morpheme parser and syntax analyzer for sequentially morphological analysis, syntax analysis, and semantic analysis by comparing the data output from the speech / text conversion device with data of the pre-storage device. And a speech analyzer to omit and substitute the data output from the semantic parser with reference to the data of the pre-storage device, and to resolve the terminology, and the data output from the speech analyzer to the data of the knowledge base storage device. Compare and convert them into subcategories of verbs, A conversation manager for performing dialogue acts, keyword in region, suitability in region, database query generation and search, and speech acts and database search results by comparing the data output from the conversation manager with the data of the speech / text converter. And a response generator for generating a sentence to be provided to the user from the region's goodness-of-fit, and provides a device and a conversation analysis method capable of acquiring a device for helping his / her learning through only a language used by a human. .

Description

Interactive Learning Aid and Its Conversation Analysis Method

The present invention relates to an interactive learning assist device and a method for analyzing the dialogue thereof, which enable a user of a computer to obtain a device for helping his / her learning through only a language used by a human being without detailed knowledge or skills of the computer.

Here, a device that helps learning is to transmit a human voice to the computer by using a microphone attached to the computer when the learner lacks knowledge in performing his or her task. Then, the computer extracts the data in the database suitable for the user's requirements or the corresponding knowledge through morphological analysis, syntactic analysis, semantic analysis, discourse analysis, and dialogue analysis of the spoken speech language and generates the knowledge in human voice language again. Means a device to pull out.

Since the 1980s, research into using computers using natural language has been continued. For example, PLANES (1978), LADDER (1978), PHLOIA1 (1980), TOA (1980), COOP (1982), TEAM (1987), IRUS (1986), TELJ (1986), LOKI (1988), JANUS ( Natural language database interface systems have been developed, and commercial systems such as INTELECT, Q & A, Natural, and Language have been developed. Existing studies are basically that the user throws a query and the system shows the result of the query. The user cannot use the previous query or the result. Therefore, the user feels more uncomfortable because the user has to enter a long sentence to make a query, and the complexity of the natural language processor increases.

Natural language processing in Korea has been studied with emphasis on morphological analysis and syntax analysis, and research on semantic processing has been studied depending on application systems such as machine translation. In these studies, meaning is used in such a way as to use semantic features or to examine valid meanings, but it does not have a formal structure that represents the overall meaning of a sentence. Recently, researches using conceptual graphs as a semantic structure have been conducted.However, in this method, parsing and semantic analysis are performed independently, so it is difficult to organize them. Parse paths using semantic information during the parsing process You won't be able to take full advantage of this. The syntax-based semantic parsing technique that combines syntactic analysis and semantic analysis is a method in which the order and method of generating the semantic structure are controlled by the syntactic information. There are advantages in reducing possible parse paths and faster processing.

Therefore, the present invention can identify and maintain the omission / substitute phenomena, conversation focus that cannot be solved in the existing natural language interface system implemented only by the response of the user's response to the user's question or the user's response to the question of the one-side system. By applying the dialogue system to the learning tool, which is implemented to solve the phenomena that occur frequently in the dialogues such as sub-dialogs and the presentation of proximity answers, general users who lack the knowledge of the computer and the internet can easily access the information and data on the Internet. An object of the present invention is to provide an interactive learning aid device and a method of analyzing the same, which can solve a problem that cannot be obtained.

Interactive learning aids according to the present invention for achieving the above object is a morphological and syntax analysis, a pre-storage device for providing data necessary for meaning and discourse analysis, and a knowledge base storage for providing data required for dialogue analysis A morphological analysis sequentially by comparing a device, a voice / text conversion device for converting a voice input through a microphone into text having the same content, and data output from the voice / text conversion device with data of the pre-storage device; A morpheme parser, a parser and a semantic parser for parsing and semantic analysis, a discourse analyzer for eliminating and substitutes the data output from the semantic interpreter with reference to the data of the pre-storage device, and the discourse analyzer The data output from the knowledge base storage device A conversation manager for converting into subcategory information of the verb compared with the data, and performing speech acts, intra-domain keywords, intra-regional suitability determination, database query word generation and retrieval, and the data output from the conversation manager. And a response generator for generating a sentence to be provided to the user from speech acts, database search results, and in-area goodness in comparison with the data of the / text conversion apparatus.

Conversation analysis method of the interactive learning aid device according to the present invention for achieving the above object in the dialogue analysis method to grasp the user's requirements from the user's speech sentence, QLF received as a result of semantic analysis (QLF) A first step of deleting or integrating the redundant information into a subcategory structure of the verb, a second step of extracting a speech act from the first step according to the meaning of the operator and the main verb, and the second step A third step of determining suitability in the area of the user's speech according to the semantic consistency of the dialogue acts and verb subcategories, and a fourth step of generating a query in the database from the subcategories of the verb from the third step. Wow; A fifth step of searching a database through a query generated from the fourth step and continuing to preserve the result by the user's ROI, or finding a specific attribute from existing database search results; and the third and fifth steps. A sixth step of generating exception processing, acknowledgment, and response utterance according to the database search result and the speech act and the suitability of utterance in the region, and the seventh step of generating additional utterance by knowledge inference from the sixth step; And an eighth step of storing the final result generated in the seventh step in a separate storage device and outputting the result through the display device of the screen / print and the ignition device of the microphone.

1 is a block diagram of an interactive learning aid device according to the present invention.

2 is a flowchart illustrating a method of generating a response in a conversation analyzer to which the present invention is applied.

3 (a) and 3 (b) are exemplary diagrams of semantic interpretation, discourse analysis, dialogue analysis results, and generated sentences.

<Description of Signs of Major Parts of Drawings>

1: microphone 2: voice / text converter

3: stemmer parser 4: parser

5: semantic analyzer 6: discourse analyzer

7: conversation manager 8: response generator

9: output device 10: prestorage device

12: knowledge base storage

Hereinafter, an embodiment of the present invention will be described in detail with reference to the accompanying drawings.

1 is a block diagram of an interactive learning aid device according to the present invention, FIG. 2 is a flowchart illustrating a method of generating a response in a conversation analyzer to which the present invention is applied, and FIGS. 3 (a) and 3 (b) are present views. An example diagram of semantic analysis, discourse analysis, dialogue analysis results, and generated sentences by a dialogue analyzer to which the invention is applied is shown.

Referring to FIG. 1, the system includes a voice / text converter 2 for converting a voice input through a microphone 1, which is a voice input device, into text having the same content, and the voice / text converter ( A morpheme interpreter (3), a parser (4) and a semantic interpreter (5) for morphological analysis, parsing and semantic analysis sequentially by comparing the data output from 2) with the data of the pre-storage device 10; Knowledge of the discourse analyzer 6 for omitting and substituting the data output from the semantic analyzer 5 with reference to the data of the pre-storage device 10 and the data output from the discourse analyzer 6 is known. Compared to the data of the base storage device 11, the information is converted into subcategory information of the verb, and from this, speech acts, keywords in a region, determination of suitability in a region, database query generation and search, etc. are performed. The conversation manager 7 for comparing the data output from the conversation manager 7 with the data of the speech / text conversion device 2 to obtain a sentence to be provided to the user from a speech act, a database search result, and a goodness of fit within a region. Response generator 8 for generating and outputting to output device 9, prestorage device 10 for providing morphological and syntax analysis, data for meaning and discourse analysis, and providing data for conversation analysis It consists of a knowledge base storage device (11). For example, suppose the learner uses the device with the following intention.

Learner's Intention: I want to get information about Impressionist painters on the Internet.

Now, assume the following dialogue scenario when the user uses the apparatus with the above intention.

<scenario>

User (1): I would like to find information about Impressionist painters.

System (1): Do you want resources for all impressionist painters?

User (2): No Tell me about each artist's work.

System (2): Please tell me your name.

User (3): Cezanne.

System (3): Cezanne's works include Bathing Man, Work A, Work B, and so on.

User (4): Where is the first piece now?

System (4): The Metropolitan Museum of New York. May I show you a picture?

User (5): Yes please show me.

User (6): Can you show other people's work?

As an example, in the above scenario, it is assumed that when a utterance such as the user 4 comes in as an input, each step will be described.

In order to respond to a user's spoken text, the system must understand the user's speech. To do that, you first need to convert the voice into text. The voice input through the microphone 1, which is the voice input device of FIG. 1, is converted into text having the same content through the voice / text converter 2.

The converted text generates an analysis result as shown in FIG. 3 through a morpheme interpreter 3, a parser 4, and a semantic interpreter 5. As shown in (a) of FIG. 3, the resulting semantic interpretation results in a substitute term or abbreviation while passing through the discourse analyzer 6. In the interpretation of the meaning of the example user's input sentence, "Where is the first piece of art stored now?", Part A of FIG. 3 (a), which corresponds to the "first piece", is the "bath" referred to in this specialty. Is converted to [named_string, bathing_man].

The conversation manager 7 generates a response from the results of the discourse analyzer 6 and the conversation analysis results from U (1) to U (N-1). As shown in FIG. 2, first, the QLF (Quasi-Logical Form) converted from the discourse analysis still contains a lot of redundant information. Therefore, it should be normalized to be convenient for subsequent analysis. In the process of normalization, unnecessary information should be eliminated or duplicated information should be unified (101). The rule of normalization to perform this process is as follows.

Omit when feature values are the default in the term structure corresponding to the noun. This is due to the characteristics of Korean, where the use of articles or numbers is not important.

2. Delete event variables used in verb phrases.

3. The form of [and, A, B] is recursively processed according to the following types.

end. A-verb phrase, B-noun phrase

A is used as a tubular sphere of B, where A represents the property of B. Thus, an object of B is created and B is assigned as the attribute value to the attribute name in the object.

I. A-noun phrase, B-noun phrase

It is a compound noun, and object and attribute values are determined according to the following eight semantics.

1) place name + institution

2) person's name + job title

3) material + stuff

4) Institution + Position

5) Category + Objects in Category

6) Uses + Tools

7) object + part

8) Property + Object

As shown in the result of the dialogue analysis of FIG. 3 (b), the QLF form after normalization is represented as an internally integrated and simplified form while maintaining the name and structure of the predicket corresponding to the main verb of a sentence.

Speech acts play an important role in dialogue analysis. The actual surface utterance is in the form of a question, but the user's intention is to provide some specific information, or it may be a mere proposition but one that demands something from the listener. In addition, since the usage of the word can also vary according to the application area, the usage of the word should be considered along with the Mood and Modal operators when analyzing speech acts. Speech acts are determined according to the following rules:

(Mood Operator, Modal Operator, Predicate)-> Speech_Act

As shown in FIG. 2, after extracting a speech act, it is necessary to determine whether the user's speech is suitable for the corresponding region (102, 103, 104, and 105). This is determined by the domain-dependent keywords extracted during the dialogue analysis, which matches the words existing in the domain. `` Expected '' for grammar / meaning, `` prenormal ''; `` possible '' for grammar / meaning, but not included in the word set in the domain; 'abnormal' for grammar / meaning. Decide on In case of 'possible', the response is generated through the exception processing unit because the probability is sufficiently sufficient, and in case of 'abnormal', the user is required to speak again by generating a confirmation speech. If 'expected', processing proceeds normally.

The normalized QLF is converted into a database query language (106). In this case, each attribute value in the created object is converted into a database query form, in which ambiguous attribute values that appear qualitatively or language-specific are mapped into concrete and quantified values. For example, [now, X20] corresponding to 'now', as shown in FIG. 3, is changed to a specific time expression.

The generation of response speech is determined by the rules as shown in Table 1 below by the user speech act, the system speech act, and the database search result (107 and 108). First, the speech act of the system by the user speech act is determined by the following rules.

1. Rules for determining system acts

ask_yesno: answer_yesno

ask_what: give_information

request_information: give_information

2. Rules for generating response utterances

System speech Database search results Response ignition type give_information N = 0 User's Condition + Database Search Result + Create Alternative N> δ User's Condition + Database Search Results + Additional Information N <δ User's Condition + Database Search Results + Question answer_yesno - yes + N (units) + `all '+ user's condition no + N (units) + 'all' + negation of user's condition no + N (units) is the negation of the user's condition + M (units) + the user's condition ask_what - Question

(N: number of objects in the database that meets user requirements, δ: any small number, here 3)

After the response speech is generated, knowledge inference and additional speech are processed (109 and 110). After generating a response through a database search from the example sentence shown in FIG. 3, the reasoning is made based on the response. Inference is performed by the following rules.

isa (obj, N1148) and num_of (filled slot)> 2-> UB (Want (see (obj)))

The reasoning of the above reasoning rule is that 'the current user is referring to an artifact of human art, and if more than two attributes are mentioned, the user believes that the user wants to see the object'.

Based on the new knowledge generated by inference, the system generates additional utterances. For example, the additional firing generation rule is as follows.

UB (Want (pred (obj)))-> [ques, modal_want, pred (obj)]

Through this rule, the system finally generates the response utterance and the additional utterance so that the utterance is the same as the system 4 of the scenario example described previously.

The present invention described above is capable of various substitutions, modifications, and changes without departing from the spirit of the present invention for those skilled in the art to which the present invention pertains. It is not limited.

As described above, according to the present invention, when the 21st century high-speed information communication network is generalized, all students will be able to receive the help of computers through conversations using people's everyday language in performing their tasks in each home. It is expected.

Claims (2)

Pre-storage to provide data for morphological and parsing, semantic and discourse analysis, A knowledge base storage device for providing data for conversation analysis; A speech / text converter for converting speech input through a microphone into text having the same content; A morpheme parser, a parser, and a semantic parser for sequentially morphological analysis, parsing and semantic analysis of data output from the speech / text conversion device with data of the pre-storage device; A discourse analyzer for omitting and substituting the data output from the semantic interpreter with reference to the data of the pre-storage device; The data output from the discourse analyzer is compared with the data of the knowledge base storage device and converted into subcategory information of the verb, and from there, a dialogue act, a keyword in a region, a suitability determination in a region, and a database query word generation and retrieval are performed. With the conversation manager, And a response generator for generating a sentence to be provided to the user from speech acts, database search results, and in-field fitness by comparing the data output from the conversation manager with the data of the speech / text converter. Auxiliary devices. In the dialogue analysis method that gradually grasps the user's requirements from the user's speech sentence, A first step of deleting or integrating excess information of QLF received as a result of semantic interpretation into subcategories of English language, A second step of extracting speech acts according to the meanings within the scope of the operator and the general language from the first step; A third step of judging suitability in the area of user speech according to the degree of meaning correspondence in the area of speech act and verb subcategory elements from the second step; A fourth step of generating a query in a database from the subcategory information of the verb from the third step; A fifth step of searching the database through the query generated from the fourth step and continuing to store the result by the user's ROI, or finding a specific attribute from the existing database search results; A sixth step of generating exception processing, confirmation speech, and response speech from the third and fifth steps according to the database search result and the speech act and the appropriateness of speech in the region, respectively; A seventh step of generating an additional utterance by knowledge inference from the sixth step, And an eighth step of storing the final result generated in the seventh step in a separate storage device and outputting the result through the display device of the screen / print and the ignition device of the microphone. Method of dialogue analysis of device.