A method for evaluating oral conversation quality in college English learning

Abstract

The present invention discloses a method for evaluating the quality of oral dialogues for college English learning, which relates to the field of data processing technology. The method comprises the following steps: constructing a dialogue knowledge database; segmenting the spoken dialogue speech into actual dialogue boxes; identifying dialogue application scenarios; obtaining semantic misunderstanding parts; obtaining a misunderstanding degree coefficient; analyzing the degree of pronunciation problems and the speed of initiating responses in the actual dialogue boxes; comprehensively obtaining a dialogue coherence index; obtaining a quality assessment value; obtaining a quality assessment value range using sample dialogue content; calculating the proportion of parts whose quality assessment values ​​exceed the quality assessment value range as a recognition ratio; and the greater the recognition ratio, the lower the quality of the spoken dialogue. By identifying the dialogue application scenarios, obtaining the semantic misunderstanding parts, obtaining the misunderstanding degree coefficient, and analyzing the degree of pronunciation problems and the speed of initiating responses, oral dialogues can be more accurately evaluated from multiple aspects.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method for evaluating the quality of spoken dialogues used in college English learning. Background Art

[0002] With the development of globalization and the deepening of communication between countries, multilingual learning is becoming increasingly important, and mastering a foreign language is extremely important. English, as a widely used foreign language, is now included in the basic curriculum of Chinese universities. However, due to the large number of students, lack of interaction, a single teaching model, and the language environment, the process of English teaching in universities has caused an imbalance between the written and oral aspects of English education. Students often achieve high scores on written tests but lack the oral skills to communicate in English.

[0003] Due to the variability of oral conversations, it is difficult to accurately evaluate the quality of oral conversations, which makes it impossible to provide targeted training for students. Summary of the Invention

[0004] In order to solve the above technical problems, a method for evaluating the quality of oral dialogues for college English learning is provided. This technical solution solves the problems raised in the above background technology.

[0005] In order to achieve the above objects, the technical solution adopted by the present invention is:

[0006] A method for evaluating the quality of spoken dialogues for college English learning, comprising:

[0007] Constructing a conversation knowledge database based on at least one sample conversation content of a native English speaker;

[0008] Acquire spoken conversation speech, and segment the spoken conversation speech into at least one actual dialog box in chronological order;

[0009] Based on the spoken conversation speech and conversation knowledge database, the conversation application scenario is identified;

[0010] Based on the dialogue application scenario, the unreasonable content is identified to obtain the semantic misunderstanding part;

[0011] Analyze the unreasonable degree of semantic misunderstanding and obtain the misunderstanding degree coefficient;

[0012] Analyze the pronunciation problem level and initiation response speed of the actual dialog box;

[0013] Based on the analysis of the proportion of semantic misunderstandings, the dialogue coherence index is obtained;

[0014] forming weights for the degree of pronunciation problems, the speed of initiating a response, and the conversation coherence index, and superimposing the degree of pronunciation problems, the speed of initiating a response, and the conversation coherence index using the corresponding weights to obtain a quality assessment value;

[0015] Use sample conversation content to obtain a quality assessment value range;

[0016] Calculate the proportion of parts whose quality assessment values ​​exceed the quality assessment value range as the recognition ratio;

[0017] The larger the recognition ratio, the lower the quality of the spoken dialogue.

[0018] Preferably, the step of constructing a conversation knowledge database based on at least one sample conversation content of a native English speaker comprises the following steps:

[0019] Based on the timbre, the sample dialogue content is segmented in chronological order to obtain at least one sample dialogue box, wherein the sample dialogue box contains spoken content of the same person, and the spoken content of adjacent sample dialogue boxes belongs to different people;

[0020] Sorting the sample dialog boxes of the sample dialog contents in chronological order to obtain a sample dialog sequence, and numbering the sample dialog boxes in the sample dialog sequence in order;

[0021] Use neural network models to build semantic recognition models;

[0022] Using the semantic recognition model, the semantics of the sample dialog in the sample dialog sequence are recognized to obtain the semantics of the sample dialog;

[0023] When the sample dialog box semantics of two sample dialog boxes are related, a correspondence relationship is established between the two sample dialog boxes;

[0024] Forming at least one associated dialogue set, where the associated dialogue set consists of sample dialogue boxes with the same sample dialogue content, and any two sample dialogue boxes in the associated dialogue set are in a corresponding relationship;

[0025] Perform semantic recognition on the sample conversation content of the generated associated conversation set to obtain sample semantics;

[0026] Retrieving at least one conversation text with the same semantic expression as the sample in the big data, obtaining the occurrence context of the conversation text, performing feature extraction on the occurrence context, and obtaining at least one application feature;

[0027] Count the number of occurrences of application features as the weight of the application features;

[0028] Accumulate the weights of the application features of the occurrence situation to obtain the judgment value;

[0029] The occurrence scenario with the largest judgment value is used as the occurrence scenario of the associated dialogue set, and the associated dialogue set and its corresponding occurrence scenarios are summarized into a dialogue knowledge database.

[0030] Preferably, the step of segmenting the spoken dialogue speech into at least one actual dialogue box in chronological order comprises the following steps:

[0031] Based on the timbre, the spoken dialogue speech is segmented in chronological order to obtain at least one actual dialog box. The actual dialog box contains the spoken content of the same person. The spoken content contained in adjacent actual dialog boxes belongs to different people. The actual dialog boxes are numbered in chronological order.

[0032] Preferably, identifying a dialogue application scenario based on the spoken dialogue speech and the dialogue knowledge database comprises the following steps:

[0033] Perform semantic recognition on the actual dialog box of the spoken conversation speech to obtain the semantics of the actual dialog box;

[0034] When the actual dialog box semantics of two actual dialog boxes are related, a correspondence relationship is established between the two actual dialog boxes;

[0035] At least one actual dialogue set is formed, where the actual dialogue set is composed of actual dialog boxes, and any two actual dialog boxes in the actual dialogue set are in a corresponding relationship;

[0036] Obtain the associated conversation set with the largest overlap with the actual conversation set, match it to the actual conversation set, and match the occurrence context of the associated conversation set to the corresponding actual conversation set;

[0037] The occurrence scenarios corresponding to the actual dialogue set of spoken dialogue speech are deduplicated and summarized to obtain the dialogue application scenario.

[0038] Preferably, the identifying of unreasonable content to obtain the semantic misunderstanding part comprises the following steps:

[0039] The associated conversation set corresponding to the occurrence scenario with the greatest overlap with the conversation application scenario is taken as the preliminary associated conversation set;

[0040] Taking one of at least one actual dialog box as the target actual dialog box;

[0041] The actual dialog set containing the target actual dialog box is used as the target actual dialog set;

[0042] The preliminary associated conversation set with the same number of elements as the target actual conversation set is used as the secondary associated conversation set;

[0043] The secondary associated conversation set with the largest overlap with the target actual conversation set is used as the target associated conversation set;

[0044] Arrange the sample dialogs in the target associated dialog set in order to obtain a first sequence, and arrange the actual dialogs in the target actual dialog set in order to obtain a second sequence;

[0045] Establish a correspondence between the sample dialog boxes and the actual dialog boxes at the same position in the first sequence and the second sequence, and use the sample dialog box corresponding to the target actual dialog box as the target sample dialog box;

[0046] Obtain the part of the target actual dialog box that is different from the target sample dialog box as the suspicious part;

[0047] Based on big data, we can obtain commonly used keywords in spoken language and classify them. Keywords of the same type have the same semantics.

[0048] Use similar keywords to replace the suspicious part and obtain all replacement situations after the suspicious part is replaced;

[0049] In each replacement case, the part of the target actual dialog box that is different from the target sample dialog box is retrieved as the unreasonable part, and the part of the target actual dialog box that is the same as the target sample dialog box is retrieved as the reasonable part;

[0050] The unreasonable part with the smallest length is regarded as the semantic misunderstanding part;

[0051] When the target actual dialog box traverses at least one actual dialog box, at least one semantic misunderstanding part is obtained.

[0052] Preferably, analyzing the unreasonable degree of the semantic misunderstanding part to obtain the misunderstanding degree coefficient comprises the following steps:

[0053] In the target sample dialog box, the reasonable part corresponding to the semantic misunderstanding part is removed to obtain the reference part;

[0054] Based on big data, the probability of two keywords appearing simultaneously is counted as the correlation between the two keywords;

[0055] Acquire all possible combinations of keywords in the semantic misunderstanding part and keywords in the reference part to obtain at least one keyword group;

[0056] The reciprocal of the average of the relevance of the keywords in at least one keyword group is taken to obtain the misunderstanding degree coefficient.

[0057] Preferably, the analysis to obtain the pronunciation problem degree and initiation response speed of the actual dialog box includes the following steps:

[0058] Compare the standard pronunciation of English words with the sample conversation content to obtain at least one sample pronunciation;

[0059] Count the minimum time interval between adjacent sample pronunciations in chronological order as the feature time;

[0060] Identify the pronunciation intervals of the content in the actual dialog box and use the points where the pronunciation interval time is greater than the characteristic time as the segmentation points;

[0061] The content in the actual dialog box between adjacent segmentation points is used as the actual word pronunciation;

[0062] Obtain words whose pronunciation differs from the actual word pronunciation by less than a preset value as suspected words, summarize the set of suspected words, and use the suspected word with the smallest pronunciation difference from the actual word in the set of suspected words as the target word. Use the target word to replace the position of the actual word pronunciation in the actual dialog box to obtain the actual change dialog box. The preset value is a parameter set based on experience;

[0063] When the semantics of the actual change dialog box is consistent with the actual dialog box, the difference between the standard pronunciation of the target word and the actual word pronunciation is counted as the degree of local problem. Otherwise, the target word is deleted from the set of suspected words and the previous step is repeated until the degree of local problem is counted.

[0064] Accumulate the local problem degrees to obtain the pronunciation problem degree;

[0065] The time intervals between adjacent actual dialog boxes are averaged to obtain the initiation reaction speed.

[0066] Preferably, the analysis based on the proportion of semantic misunderstandings to comprehensively obtain the dialogue coherence index includes the following steps:

[0067] The proportion of semantic misunderstanding in the spoken conversation speech is multiplied by the misunderstanding degree coefficient of the semantic misunderstanding part and then added together to obtain the conversation coherence index.

[0068] Preferably, forming the weights of the pronunciation problem degree, the initiation response speed and the conversation coherence index comprises the following steps:

[0069] Based on the analytic hierarchy process, weights for the degree of pronunciation problems, speed of initiation response, and conversation coherence index were formed.

[0070] Preferably, obtaining a quality assessment value range using the sample conversation content comprises the following steps:

[0071] Obtaining a quality evaluation value of the sample conversation content in the same manner as the quality evaluation value of the spoken conversation speech;

[0072] The maximum and minimum values ​​of the quality evaluation values ​​of the sample conversation contents are used as endpoints to form a quality evaluation value range.

[0073] Compared with the prior art, the present invention has the following beneficial effects:

[0074] By identifying the dialogue application scenario, the semantic misunderstanding part, the misunderstanding degree coefficient, and analyzing the degree of pronunciation problems and the speed of initiating response, the important situations in oral dialogue are comprehensively considered, and the errors in semantic understanding and pronunciation problems are focused on. Then, oral dialogue can be evaluated more accurately from multiple aspects, and targeted training can be carried out for students in a timely manner according to the situation of the dialogue. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 Schematic diagram of the flow of the oral dialogue quality assessment method for college English learning of the present invention;

[0076] Figure 2 A schematic diagram of a process for constructing a conversation knowledge database based on at least one sample conversation content of a native English speaker according to the present invention;

[0077] Figure 3 This is a flow chart of the present invention for identifying a dialogue application scenario based on spoken dialogue speech and a dialogue knowledge database;

[0078] Figure 4 A schematic diagram of the flow of identifying unreasonable content and obtaining semantic misunderstandings in the present invention;

[0079] Figure 5 A flow chart of analyzing the unreasonable degree of semantic misunderstanding to obtain a misunderstanding degree coefficient according to the present invention;

[0080] Figure 6 A flow chart of the present invention for analyzing and obtaining the degree of pronunciation problems and the speed of initiating response of an actual dialog box;

[0081] Figure 7 Schematic diagram of the flow of obtaining a quality assessment value range using sample conversation content according to the present invention. DETAILED DESCRIPTION

[0082] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0083] Reference Figure 1 As shown, a method for evaluating the quality of oral dialogue for college English learning includes:

[0084] Constructing a conversation knowledge database based on at least one sample conversation content of a native English speaker;

[0085] Acquire spoken conversation speech, and segment the spoken conversation speech into at least one actual dialog box in chronological order;

[0086] Based on the spoken conversation speech and conversation knowledge database, the conversation application scenario is identified;

[0087] Based on the dialogue application scenario, the unreasonable content is identified to obtain the semantic misunderstanding part;

[0088] Analyze the unreasonable degree of semantic misunderstanding and obtain the misunderstanding degree coefficient;

[0089] Analyze the pronunciation problem level and initiation response speed of the actual dialog box;

[0090] Based on the analysis of the proportion of semantic misunderstandings, the dialogue coherence index is obtained;

[0091] forming weights for the degree of pronunciation problems, the speed of initiating a response, and the conversation coherence index, and superimposing the degree of pronunciation problems, the speed of initiating a response, and the conversation coherence index using the corresponding weights to obtain a quality assessment value;

[0092] Use sample conversation content to obtain a quality assessment value range;

[0093] Calculate the proportion of parts whose quality assessment values ​​exceed the quality assessment value range as the recognition ratio;

[0094] The larger the recognition ratio, the lower the quality of the spoken dialogue.

[0095] The most important part of oral quality assessment is the evaluation of the rationality of the semantic understanding of the dialogue. In some cases, although there are no grammatical errors and no obvious problems with the content of the dialogue, it is completely inconsistent with the current situation, which is actually irrelevant to the question. Therefore, it is necessary to identify the extent of this situation. The second is to identify pronunciation problems. Since it is not clear which word the student is saying, there will be greater obstacles to identifying pronunciation problems. Therefore, corresponding steps need to be set to solve the problem. The speed of response actually reflects the student's mastery of the language, so it also needs to be taken into consideration, but it is relatively easy to identify.

[0096] Reference Figure 2 As shown, constructing a conversation knowledge database based on at least one sample conversation content of a native English speaker includes the following steps:

[0097] Based on the timbre, the sample dialogue content is segmented in chronological order to obtain at least one sample dialogue box, wherein the sample dialogue box contains spoken content of the same person, and the spoken content of adjacent sample dialogue boxes belongs to different people;

[0098] Sorting the sample dialog boxes of the sample dialog contents in chronological order to obtain a sample dialog sequence, and numbering the sample dialog boxes in the sample dialog sequence in order;

[0099] Use neural network models to build semantic recognition models;

[0100] Using the semantic recognition model, the semantics of the sample dialog in the sample dialog sequence are recognized to obtain the semantics of the sample dialog;

[0101] When the sample dialog box semantics of two sample dialog boxes are related, a correspondence relationship is established between the two sample dialog boxes;

[0102] Forming at least one associated dialogue set, where the associated dialogue set consists of sample dialogue boxes with the same sample dialogue content, and any two sample dialogue boxes in the associated dialogue set are in a corresponding relationship;

[0103] Perform semantic recognition on the sample conversation content of the generated associated conversation set to obtain sample semantics;

[0104] Retrieving at least one conversation text with the same semantic expression as the sample in the big data, obtaining the occurrence context of the conversation text, performing feature extraction on the occurrence context, and obtaining at least one application feature;

[0105] Count the number of occurrences of application features as the weight of the application features;

[0106] Accumulate the weights of the application features of the occurrence situation to obtain the judgment value;

[0107] The occurrence scenario with the largest judgment value is used as the occurrence scenario of the associated dialogue set, and the associated dialogue set and its corresponding occurrence scenarios are summarized into a dialogue knowledge database.

[0108] The dialogue knowledge database mainly forms the corresponding occurrence context for the related dialogue set. Subsequently, the context can be judged based on the occurrence context, and then the related dialogue set corresponding to the same context can be called. The relationship between the related dialogue set is similar to the context, because each dialogue sentence is not just a response to the previous sentence, but may also be a response to the previous sentences. Therefore, it is necessary to identify related dialogues and summarize them for consideration.

[0109] Segmenting the spoken dialogue speech into at least one actual dialog box in chronological order comprises the following steps:

[0110] Based on the timbre, the spoken dialogue speech is segmented in chronological order to obtain at least one actual dialog box. The actual dialog box contains the spoken content of the same person. The spoken content contained in adjacent actual dialog boxes belongs to different people. The actual dialog boxes are numbered in chronological order.

[0111] Spoken conversations are conducted by several people, each of whom communicates in turn. Therefore, segmenting the conversations based on timbre and obtaining the actual dialog box can facilitate subsequent analysis.

[0112] Reference Figure 3 As shown, based on the spoken conversation speech and the conversation knowledge database, identifying the conversation application scenario includes the following steps:

[0113] Perform semantic recognition on the actual dialog box of the spoken conversation speech to obtain the semantics of the actual dialog box;

[0114] When the actual dialog box semantics of two actual dialog boxes are related, a correspondence relationship is established between the two actual dialog boxes;

[0115] At least one actual dialogue set is formed, where the actual dialogue set is composed of actual dialog boxes, and any two actual dialog boxes in the actual dialogue set are in a corresponding relationship;

[0116] Obtain the associated conversation set with the largest overlap with the actual conversation set, match it to the actual conversation set, and match the occurrence context of the associated conversation set to the corresponding actual conversation set;

[0117] The occurrence scenarios corresponding to the actual dialogue set of spoken dialogue speech are deduplicated and summarized to obtain the dialogue application scenario.

[0118] When conducting semantic analysis, the first thing to do is to determine the context of the conversation. Otherwise, without the constraints of the context, it is easy to make analysis errors, because different conversations may be reasonable or unreasonable in different contexts. Therefore, when the context cannot be determined, it is not convenient to analyze the rationality of the conversation.

[0119] Reference Figure 4 As shown, identifying unreasonable content and obtaining semantic misunderstandings includes the following steps:

[0120] The associated conversation set corresponding to the occurrence scenario with the greatest overlap with the conversation application scenario is taken as the preliminary associated conversation set;

[0121] Taking one of at least one actual dialog box as the target actual dialog box;

[0122] The actual dialog set containing the target actual dialog box is used as the target actual dialog set;

[0123] The preliminary associated conversation set with the same number of elements as the target actual conversation set is used as the secondary associated conversation set;

[0124] The secondary associated conversation set with the largest overlap with the target actual conversation set is used as the target associated conversation set;

[0125] Arrange the sample dialogs in the target associated dialog set in order to obtain a first sequence, and arrange the actual dialogs in the target actual dialog set in order to obtain a second sequence;

[0126] Establish a correspondence between the sample dialog boxes and the actual dialog boxes at the same position in the first sequence and the second sequence, and use the sample dialog box corresponding to the target actual dialog box as the target sample dialog box;

[0127] Obtain the part of the target actual dialog box that is different from the target sample dialog box as the suspicious part;

[0128] Based on big data, we can obtain commonly used keywords in spoken language and classify them. Keywords of the same type have the same semantics.

[0129] Use similar keywords to replace the suspicious part and obtain all replacement situations after the suspicious part is replaced;

[0130] In each replacement case, the part of the target actual dialog box that is different from the target sample dialog box is retrieved as the unreasonable part, and the part of the target actual dialog box that is the same as the target sample dialog box is retrieved as the reasonable part;

[0131] The unreasonable part with the smallest length is regarded as the semantic misunderstanding part;

[0132] When the target actual dialog box traverses at least one actual dialog box, at least one semantic misunderstanding part is obtained.

[0133] Since the situations of daily conversations are relatively limited, all possibilities can be approximated by constructing a conversation knowledge database. Although it is impossible to include all possibilities, basically, actual conversations can find approximately corresponding situations in the conversation knowledge database, so the target-related conversation set is obtained from it. The structure of the target-related conversation set is the most similar to that of the target actual conversation set, and the scenario is also the closest. Since the elements in the target-related conversation set and the target actual conversation set are arranged in chronological order, the target sample dialog box with the same relative position can be found for comparison with the target actual dialog box, and then the unreasonable part can be obtained. However, due to the existence of synonyms, replacement comparison is required to finally determine the semantic misunderstanding part.

[0134] Reference Figure 5 As shown, analyzing the unreasonable degree of the semantic misunderstanding part and obtaining the misunderstanding degree coefficient includes the following steps:

[0135] In the target sample dialog box, the reasonable part corresponding to the semantic misunderstanding part is removed to obtain the reference part;

[0136] Based on big data, the probability of two keywords appearing simultaneously is counted as the correlation between the two keywords;

[0137] Acquire all possible combinations of keywords in the semantic misunderstanding part and keywords in the reference part to obtain at least one keyword group;

[0138] The reciprocal of the average of the relevance of the keywords in at least one keyword group is taken to obtain the misunderstanding degree coefficient.

[0139] The mean of the correlation is an evaluation of the correlation. Therefore, when it is reciprocated, it is an evaluation of the non-correlation. Therefore, it can be understood as the misunderstanding degree coefficient.

[0140] Reference Figure 6 As shown, analyzing the pronunciation problem level and initiation response speed of the actual dialog box includes the following steps:

[0141] Compare the standard pronunciation of English words with the sample conversation content to obtain at least one sample pronunciation;

[0142] Count the minimum time interval between adjacent sample pronunciations in chronological order as the feature time;

[0143] Identify the pronunciation intervals of the content in the actual dialog box and use the points where the pronunciation interval time is greater than the characteristic time as the segmentation points;

[0144] The content in the actual dialog box between adjacent segmentation points is used as the actual word pronunciation;

[0145] Obtain words whose pronunciation differs from the actual word pronunciation by less than a preset value as suspected words, summarize the set of suspected words, and use the suspected word with the smallest pronunciation difference from the actual word in the set of suspected words as the target word. Use the target word to replace the position of the actual word pronunciation in the actual dialog box to obtain the actual change dialog box. The preset value is a parameter set based on experience;

[0146] When the semantics of the actual change dialog box is consistent with the actual dialog box, the difference between the standard pronunciation of the target word and the actual word pronunciation is counted as the degree of local problem. Otherwise, the target word is deleted from the set of suspected words and the previous step is repeated until the degree of local problem is counted.

[0147] Accumulate the local problem degrees to obtain the pronunciation problem degree;

[0148] The time intervals between adjacent actual dialog boxes are averaged to obtain the initiation reaction speed.

[0149] The preset value is set based on experience. When the pronunciation difference does not exceed the preset value, the two pronunciations can be considered to be the same pronunciation. This can be obtained based on empirical testing;

[0150] Since the pronunciation of the actual word may not be standard, it is necessary to determine the actual corresponding word, and then compare the actual word pronunciation with the actual corresponding word to identify the pronunciation problem.

[0151] Based on the analysis of the proportion of semantic misunderstandings, the dialogue coherence index is obtained by the following steps:

[0152] The proportion of semantic misunderstanding in the spoken conversation speech is multiplied by the misunderstanding degree coefficient of the semantic misunderstanding part and then added together to obtain the conversation coherence index.

[0153] The weights for the pronunciation problem level, initiation response speed, and conversation coherence index are formed by the following steps:

[0154] Based on the analytic hierarchy process, weights for the degree of pronunciation problems, speed of initiation response, and conversation coherence index were formed.

[0155] Reference Figure 7 As shown, obtaining the quality assessment value range using the sample conversation content includes the following steps:

[0156] Obtaining a quality evaluation value of the sample conversation content in the same manner as the quality evaluation value of the spoken conversation speech;

[0157] The maximum and minimum values ​​of the quality evaluation values ​​of the sample conversation contents are used as endpoints to form a quality evaluation value range.

[0158] Furthermore, the present solution also proposes a storage medium having a computer-readable program stored thereon, and when the computer-readable program is called, the above-mentioned oral dialogue quality assessment method for college English learning is executed.

[0159] It is understandable that the storage medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid state disk (SSD).

[0160] In summary, the advantages of the present invention are: by identifying the dialogue application scenario, obtaining the semantic misunderstanding part, obtaining the misunderstanding degree coefficient, and analyzing the pronunciation problem degree and the response speed, the important situations in the oral dialogue are comprehensively considered, and the errors in semantic understanding and pronunciation problems are focused on, so as to more accurately evaluate the oral dialogue from multiple aspects, and provide targeted training to students in a timely manner according to the situation of the dialogue.

[0161] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for evaluating the quality of spoken dialogues in college English learning, characterized in that: include: Constructing a conversation knowledge database based on at least one sample conversation content of a native English speaker; Acquire spoken conversation speech, and segment the spoken conversation speech into at least one actual dialog box in chronological order; Based on the spoken conversation speech and conversation knowledge database, the conversation application scenario is identified; Based on the dialogue application scenario, the unreasonable content is identified to obtain the semantic misunderstanding part; Analyze the unreasonable degree of semantic misunderstanding and obtain the misunderstanding degree coefficient; Analyze the pronunciation problem level and initiation response speed of the actual dialog box; Based on the analysis of the proportion of semantic misunderstandings, the dialogue coherence index is obtained; forming weights for the degree of pronunciation problems, the speed of initiating a response, and the conversation coherence index, and superimposing the degree of pronunciation problems, the speed of initiating a response, and the conversation coherence index using the corresponding weights to obtain a quality assessment value; Use sample conversation content to obtain a quality assessment value range; Calculate the proportion of parts whose quality assessment values ​​exceed the quality assessment value range as the recognition ratio; The larger the recognition ratio, the lower the quality of the spoken dialogue.

2. The method for evaluating the quality of spoken dialogue for college English learning according to claim 1, wherein: The step of constructing a conversation knowledge database based on at least one sample conversation content of a native English speaker comprises the following steps: Based on the timbre, the sample dialogue content is segmented in chronological order to obtain at least one sample dialogue box, wherein the sample dialogue box contains spoken content of the same person, and the spoken content of adjacent sample dialogue boxes belongs to different people; Sorting the sample dialog boxes of the sample dialog contents in chronological order to obtain a sample dialog sequence, and numbering the sample dialog boxes in the sample dialog sequence in order; Use neural network models to build semantic recognition models; Using the semantic recognition model, the semantics of the sample dialog in the sample dialog sequence are recognized to obtain the semantics of the sample dialog; When the sample dialog box semantics of two sample dialog boxes are related, a correspondence relationship is established between the two sample dialog boxes; Forming at least one associated dialogue set, where the associated dialogue set consists of sample dialogue boxes with the same sample dialogue content, and any two sample dialogue boxes in the associated dialogue set are in a corresponding relationship; Perform semantic recognition on the sample conversation content of the generated associated conversation set to obtain sample semantics; Retrieving at least one conversation text with the same semantic expression as the sample in the big data, obtaining the occurrence context of the conversation text, performing feature extraction on the occurrence context, and obtaining at least one application feature; Count the number of occurrences of application features as the weight of the application features; Accumulate the weights of the application features of the occurrence situation to obtain the judgment value; The occurrence scenario with the largest judgment value is used as the occurrence scenario of the associated dialogue set, and the associated dialogue set and its corresponding occurrence scenarios are summarized into a dialogue knowledge database.

3. The method for evaluating the quality of spoken dialogue for college English learning according to claim 2, wherein: The step of segmenting the spoken dialogue speech into at least one actual dialogue box in chronological order comprises the following steps: Based on the timbre, the spoken dialogue speech is segmented in chronological order to obtain at least one actual dialog box. The actual dialog box contains the spoken content of the same person. The spoken content contained in adjacent actual dialog boxes belongs to different people. The actual dialog boxes are numbered in chronological order.

4. The method for evaluating the quality of spoken dialogue for college English learning according to claim 3, wherein: The process of identifying a dialogue application scenario based on the spoken dialogue speech and the dialogue knowledge database comprises the following steps: Perform semantic recognition on the actual dialog box of the spoken conversation speech to obtain the semantics of the actual dialog box; When the actual dialog box semantics of two actual dialog boxes are related, a correspondence relationship is established between the two actual dialog boxes; At least one actual dialogue set is formed, where the actual dialogue set is composed of actual dialog boxes, and any two actual dialog boxes in the actual dialogue set are in a corresponding relationship; Obtain the associated conversation set with the largest overlap with the actual conversation set, match it to the actual conversation set, and match the occurrence context of the associated conversation set to the corresponding actual conversation set; The occurrence scenarios corresponding to the actual dialogue set of spoken dialogue speech are deduplicated and summarized to obtain the dialogue application scenario.

5. The method for evaluating the quality of spoken dialogue for college English learning according to claim 4, wherein: The identification of unreasonable content and obtaining the semantic misunderstanding part includes the following steps: The associated conversation set corresponding to the occurrence scenario with the greatest overlap with the conversation application scenario is taken as the preliminary associated conversation set; Taking one of at least one actual dialog box as the target actual dialog box; The actual dialog set containing the target actual dialog box is used as the target actual dialog set; The preliminary associated conversation set with the same number of elements as the target actual conversation set is used as the secondary associated conversation set; The secondary associated conversation set with the largest overlap with the target actual conversation set is used as the target associated conversation set; Arrange the sample dialogs in the target associated dialog set in order to obtain a first sequence, and arrange the actual dialogs in the target actual dialog set in order to obtain a second sequence; Establish a correspondence between the sample dialog boxes and the actual dialog boxes at the same position in the first sequence and the second sequence, and use the sample dialog box corresponding to the target actual dialog box as the target sample dialog box; Obtain the part of the target actual dialog box that is different from the target sample dialog box as the suspicious part; Based on big data, we can obtain commonly used keywords in spoken language and classify them. Keywords of the same type have the same semantics. Use similar keywords to replace the suspicious part and obtain all replacement situations after the suspicious part is replaced; In each replacement case, the part of the target actual dialog box that is different from the target sample dialog box is retrieved as the unreasonable part, and the part of the target actual dialog box that is the same as the target sample dialog box is retrieved as the reasonable part; The unreasonable part with the smallest length is regarded as the semantic misunderstanding part; When the target actual dialog box traverses at least one actual dialog box, at least one semantic misunderstanding part is obtained.

6. The method for evaluating the quality of spoken dialogue for college English learning according to claim 5, wherein: Analyzing the unreasonable degree of the semantic misunderstanding part to obtain the misunderstanding degree coefficient includes the following steps: In the target sample dialog box, the reasonable part corresponding to the semantic misunderstanding part is removed to obtain the reference part; Based on big data, the probability of two keywords appearing simultaneously is counted as the correlation between the two keywords; Acquire all possible combinations of keywords in the semantic misunderstanding part and keywords in the reference part to obtain at least one keyword group; The reciprocal of the average of the relevance of the keywords in at least one keyword group is taken to obtain the misunderstanding degree coefficient.

7. A method for evaluating the quality of spoken dialogue for college English learning according to claim 6, characterized in that: The analysis to obtain the pronunciation problem degree and initiation response speed of the actual dialog box includes the following steps: Compare the standard pronunciation of English words with the sample conversation content to obtain at least one sample pronunciation; Count the minimum time interval between adjacent sample pronunciations in chronological order as the feature time; Identify the pronunciation intervals of the content in the actual dialog box and use the points where the pronunciation interval time is greater than the characteristic time as the segmentation points; The content in the actual dialog box between adjacent segmentation points is used as the actual word pronunciation; Obtain words whose pronunciation differs from the actual word pronunciation by less than a preset value as suspected words, summarize the set of suspected words, and use the suspected word with the smallest pronunciation difference from the actual word in the set of suspected words as the target word. Use the target word to replace the position of the actual word pronunciation in the actual dialog box to obtain the actual change dialog box. The preset value is a parameter set based on experience; When the semantics of the actual change dialog box is consistent with the actual dialog box, the difference between the standard pronunciation of the target word and the actual word pronunciation is counted as the degree of local problem. Otherwise, the target word is deleted from the set of suspected words and the previous step is repeated until the degree of local problem is counted. Accumulate the local problem degrees to obtain the pronunciation problem degree; The time intervals between adjacent actual dialog boxes are averaged to obtain the initiation reaction speed.

8. The method for evaluating the quality of spoken dialogue for college English learning according to claim 7, wherein: The analysis of the proportion of semantic misunderstandings to obtain a comprehensive dialogue coherence index includes the following steps: The proportion of semantic misunderstanding in the spoken conversation speech is multiplied by the misunderstanding degree coefficient of the semantic misunderstanding part and then added together to obtain the conversation coherence index.

9. The method for evaluating the quality of spoken dialogue for college English learning according to claim 8, wherein: The weighting of the pronunciation problem degree, the initiation response speed and the conversation coherence index comprises the following steps: Based on the analytic hierarchy process, weights for the degree of pronunciation problems, speed of initiation response, and conversation coherence index were formed.

10. The method for evaluating the quality of spoken dialogue for college English learning according to claim 9, characterized in that: The method of obtaining a quality assessment value range using the sample conversation content includes the following steps: Obtaining a quality evaluation value of the sample conversation content in the same manner as the quality evaluation value of the spoken conversation speech; The maximum and minimum values ​​of the quality evaluation values ​​of the sample conversation contents are used as endpoints to form a quality evaluation value range.

Images