RU2595616C1 - Method for prediction of efficiency of speech action of fragments of discourse in different languages - Google Patents

Abstract

FIELD: data processing; psychology.

SUBSTANCE: invention relates to means for prediction of effectiveness of discourse speech fragments effect at different languages. Collected parameters which can affect recipient; on one of analysed languages collected fragments of discourse containing selected parameters; extracted extra-linguistic factors (ELF) affecting effectiveness of speech effect (SEE) fragments in analysed languages and level of their effect in selected fragments by modification of these fragments for experiment, independent of extraneous factors; levels are fragment volumes, developed software and make online application, including modified fragments with selected parameters control added fragment; converted online application on investigated languages with preservation of potential speech action; online language carriers questionnaire survey is performed; correlation analysis on Pearson coefficient for each language separately to make regression equation; comparative analysis of SEE is carried out for fragments with different parameters on analysed languages; quantitative parameters contribution in value of SEE is determined and they are ranked.

EFFECT: prediction of SEE discourse fragments at different languages.

2 cl, 4 dwg, 7 tbl

Description

The invention relates to the field of linguistics, psycholinguistics and sociolinguistics and can be used in the compilation of speeches of various nature, including designed for the mass addressee, in the formation of patterns, concepts, and other mental formations of thinking, propaganda, advertising, the media, diplomatic communication, business communication, etc.

Definitions of key concepts.

Discourse is a complex-structured dynamic communicative-sign system that interacts with objective reality according to the dialectical principle; having six main plans: intentional, relevant, virtual, contextual, psychological and "sedimentary".

A discourse segment is a part of a discourse of at least 30000-40000 words

A discourse fragment is a combination of several discourse segments, the boundaries of which can be unions, punctuation marks or attribute sentences.

Control fragment - a fragment of discourse, the respondent’s response to which is supposed to be obvious by the researcher; used to filter the sample of respondents according to a specific criterion.

Extralinguistic factor (ELF) is a property or characteristic of extralinguistic social reality that causes changes in the language of both global and private nature.

The effectiveness of speech exposure (ERV) is the effective achievement of the regulation of recipient behavior, carried out using speech addressed to him (oral, written, through the media, etc.) and with the lowest possible costs.

Speech Impact Potential (RV) - the ability of linguistic means to exert a speech effect.

The standard of the norm of speech exposure is a landmark measure that serves to determine the deviation of a certain value of RV from the initial value of RV, calculated for the fragment with the smallest RV.

Addressee - sender; which in the processes of social communication through the transmission of messages tends to cause a certain behavior in the recipient.

The recipient is the subject of social communication, perceiving the message addressed to him.

A rhetorical device is a deviation from the language norm motivated by the conditions of verbal communication.

Density of rhetorical devices - a value that describes the presence of a certain number of rhetorical devices in a certain volume of discourse.

Nonparameter is a characteristic of a linguistic object, which is the absence of parameters in it provided for by the research hypothesis.

Cognitive approach - an approach based on the provisions of cognitive psychology and providing for the support of the principle of consciousness, taking into account various cognitive styles and strategies associated with cognition and thinking.

Semiotics is the science of signs and sign systems, sign (using signs) behavior and sign - linguistic and non-linguistic - communication.

A dummy variable is a qualitative variable that takes values 0 and 1 (binary), and a larger number of values (multi-valued), which is included in the econometric model to take into account the influence of qualitative features and events on the variable being explained.

The Pearson coefficient is a measure of the tightness of the relationship between two signs in the presence of a linear relationship between them.

The multiple linear regression equation is a mathematical equation that establishes a linear relationship between the resulting attribute and a number of independent parameters.

An approximate regression equation is a mathematical equation that establishes a linear relationship between the resulting attribute and a number of independent parameters; however, the coefficients of the equation can vary in a limited range depending on the representativeness and size of the sample.

The problem of finding optimal, most effective forms of expression of thought, forms of introducing a certain concept of world order into the human mind and forms of influence on the latter is one of the most relevant today. There are many linguistic and psycholinguistic works on describing and evaluating ERV as domestic ones (O.S. Issers, V.E. Chernyavskaya, K.V. Nikitina, Yu.I. Plahotnaya, S.A. Vinogradova, M.R. Zheltukhina , I.A. Skripak et al.) And foreign (T.A. van Dijk, W. Ladov, D. Franchel, G. Kress, T. Zheng, P. DiMaggio, P. Catellani, M. Dertolotti et al.), however, in all works, discourse and, in particular, the ERV of its fragments are studied using proper philological methods (content analysis, cognitive analysis, comparative, comparative analysis, ranking, etc.) or psycholinguistic (disc with analysis and others.).

The main directions in the study of speech exposure (RV) include studies of RV as a psychological, cognitive and communicative phenomenon.

The psychological approach considers the RV as certain manipulations, the purpose of which is to change a person’s behavior and thinking through motivation or demotivation, by satisfying basic needs: physiological, self-preservation, love, respect, self-affirmation. This approach evaluates the ERV through changes in human thinking and behavior, abstract criteria.

The disadvantages of the approach are the subjectivity of the ERV assessment due to the different assessment of changes in thinking and human behavior; unsystematic set of criteria for evaluating ERV, their lack of specificity, inaccuracy and dependence on various subject and non-subject ELFs; the impossibility of any quantitative assessment of the ERV; the impossibility of comparing different sections of the discourse, if it is not known how much and how the thinking or behavior of a person has changed after perceiving a fragment of the discourse (listening, viewing, reading, etc.)

The cognitive approach considers RV as a modification of the cognitive model of the world of the recipient, perceiving a fragment of discourse. Modification occurs as a result of intensification (for example, exaggeration of positive characteristics) or understatement (default, concealment of negative characteristics). These two strategies of real estate are implemented using various transformations: annihilating (for example, half-truths), finging (for example, “raising teeth”), indefinite (for example, changing the level of abstraction of phenomena or objects). All strategies are based on the introduction of new knowledge into the cognitive model of the world of the recipient or inducing the representation of existing knowledge in a distorted form.

The disadvantages of the approach are the inability to determine the degree of reliability or distortion of the cognitive model of the world of the recipient before and after the perception of the last fragment of the discourse; the impossibility of determining or accurately determining how much the recipient has acquired knowledge and how much has lost; the subjectivity of evaluating one or another fragment of the discourse, such as half-truths or changes in the level of abstraction: one researcher believes that the level of abstraction has changed, another believes that not, i.e. there are no criteria for the unambiguous determination of parameters leading to a change in the cognitive model of the recipient's world.

The communicative approach to real estate is connected with semiotics and the model of a communicative act, predetermining the potential of real estate and the conditions for its effective implementation. According to the approach, the ERV increases with a change in the mental image of the addressee, the image of the recipient, when choosing a message code, a communication channel, and in the presence of feedback. The communicative approach sees the basis of the ERV in the asymmetry of the linguistic sign.

However, the disadvantages of this approach are the impossibility of objective and accurate diagnosis of any changes in the mental image of the addressee, the image of the recipient: for

this is not developed special technical equipment. There is only general information about the choice of a message code and a communication channel: a certain message code or communication channel has the highest ERV (according to most scientists). Accordingly, it is impossible to evaluate, compare the ERV of various fragments of discourse, and even more so, in exact terms within the framework of this approach. In addition, thousands of fragments of discourse fall into the class of a certain message code and communication channel, even with a certain mental image of the addressee and the recipient - in connection with which, the researcher is forced to consider their ERV to be the same. In turn, in order to distinguish such classes, the researcher must take into account all combinations of typical images of the addressee and the recipient, message codes, communication channels and types of feedback, which is a laborious job that can take decades.

None of the known studies of RS makes an attempt to objectively evaluate its effectiveness, let alone measure it. All works find indirect evidence (for example, from the implications of texts, based on common sense and logic) of the presence or absence of real estate. If this is possible to do more or less objectively - although the indirect methods of studying the RS through exclusively the content of the discourse do not take into account a large number of contributing factors that may affect the results of the study - then comparing fragments of the discourse and other linguistic data among themselves, i.e. the assessment of their ERV relative to each other in such works is subjective, since it is impossible only by philological methods and by calculating the frequency and percentage values to estimate the difference between the ERV of one fragment and another. In addition, if there are no exact values, each researcher may have his own opinion about how strongly this or that fragment affects; the opinion of the researcher may depend on his belonging to any linguistic school, direction, culture (Slavic, Western, Eastern, etc.), etc., especially if the study is carried out on the basis of languages from different language groups. Consequently, such a study will be subjective to a large extent.

The main way to increase the objectivity and accuracy of ERV research lies through the direct receipt of data from a large number of potential recipients to whom RV can be sent, using online questioning, converting qualitative parameters and indicators into quantitative and statistical processing. Recipients are native speakers, since it is on them that RV is directed.

An object of the invention is the accurate measurement, comparison and prediction of the ERV of a discourse fragment in any language by converting qualitative parameters into quantitative ones, including by assigning values to dummy variables according to the hypotheses put forward and introducing a quantitative assessment of the degree of persuasiveness of the fragment; the possibility of one-person low-cost

research based on the responses of an unlimited number of respondents (online questionnaires) in a short time and automatic statistical processing.

The problem is achieved in that the method for predicting the effectiveness of speech exposure (ERV) of a fragment includes selecting ERV parameters in the languages studied, selecting fragments of discourse with these parameters, determining ELFs that affect ERV, leveling the effect of these ELFs in selected fragments, compiling online questionnaires and conducting online questionnaires in the languages studied (translation is carried out while preserving the potential of RS), conducting a correlation and comparative analysis of the results of online questioning with representation of the regression equation to predict the ERV depending on the presence and quantity of ERV parameters ranking parameter data according to the standardized values Beta-coefficients.

A definitive conclusion about how effective a particular fragment is from the point of view of speech exposure is formulated in accordance with the compiled regression equation if the Pearson coefficient for this model is not less than 0.1, which indicates the presence of a connection between the parameters included in the model and the EDS.

The method allows you to objectively assess and accurately measure the ERV of a discourse fragment by directly assessing the persuasiveness of a fragment by native speakers on a 5-point scale by online questioning and by converting qualitative parameters into quantitative ones, including assigning values to dummy variables.

As the parameters of the ERV, we consider all possible rhetorical techniques in the discourse that can influence the ERV of a fragment: metaphor, comparison, repetition, stylistic contrast, non-stylistic contrast, epithet, irony / satire, rhetorical question, exaggeration / understatement, pun (wordplay), neologism, allusion, rearrangement of parts of a word / sentence, replacing a word / structure with another word / structure, personification, zevgma, gradation, density of rhetorical devices (18 parameters).

At the same time, those that are a type or subspecies of the parameters highlighted above are excluded from the array of parameters. So, metonymy and synecdoch are referred to the “metaphor” parameter, since all three rhetorical devices are based on the similarity of the sign (s) of two objects of reality and the transfer of this sign to another object.

Additionally, other parameters specific to specific languages may be included in the analysis. In this case, all other procedures of the method of measuring the electric propulsion are the same.

Unfilled questionnaires are excluded from the array of received online profiles. do not give a general picture of the answers, as well as those online questionnaires that were filled out over a period of at least 1.5 minutes and no more than 10 minutes. Since the task corresponds to measuring only the electric propulsion, then automatically or thoughtlessly filling out online questionnaires (as if they were filled out

within a period of less than 1.5 minutes) or, conversely, a deeper understanding of the essence of the fragment (as if they were filled over a period of more than 10 minutes), rather than a quick reaction of the respondent with a subsequent click on one of the categories 5- A scale of answers can add up to a biased picture of the results, especially if the number of such online profiles exceeds 5% of the total. The average time for filling out online profiles is 4-6 minutes. and is calculated based on the number of fragments in the online application and their length. The most optimal number of fragments for linguistic Internet surveys is 8, and the length of each fragment is 300 characters.

In order to increase the accuracy and objectivity of the claimed method, an indirect analysis of the arithmetic mean values of the ERV is an indirect confirmation of the correctness of the compiled regression equations for the ERW, and the equation is verified by analyzing the proportions of the ERV values calculated using the equation and by analyzing the arithmetic mean values of the respondents' real answers to certain fragments randomly selected . If the difference between the values obtained by dividing the values obtained using the equation by the values obtained using the analysis of arithmetic mean values does not exceed 0.1, then the regression equation for this language is considered accurate; if this difference is in the range of 0.1-0.3 - then approximate; if the difference is more than 0.3, then it is inaccurate. Obtaining an approximate equation can be caused by the influence of factors that were not taken into account in the previous steps of the measurement method.

However, the recognition of the equation approximate does not mean failure in terms of the overall assessment of the ERV and comparison of the parameters of the ERV. Reliable correlation coefficients calculated to test the hypotheses put forward are sufficient grounds for conclusions about partial confirmation of hypotheses and the presence of trends in the languages being studied. Also, these coefficients for different languages can be compared with each other.

The invention consists in a method for measuring, comparing and predicting an ERV fragment of a discourse in different languages by means of an online questionnaire and a correlation and comparative analysis of its results; special software is being developed for conducting online questionnaires, including automatic recording of respondents' answers in a specialized database suitable for subsequent filtering on the selected EW parameters, as well as sociological parameters: gender, age, education, and for subsequent automatic statistical processing in specialized software packages Microsoft Excel, STATISTICA, UNISTAT 6.5, MedCalc, SPSS 20, SAS et al .; in this case, statistical processing provides data suitable for a comparative analysis of the ERV values and its parameters in the studied languages, namely: Pearson correlation coefficients and standardized Beta coefficients are calculated, and

regression equations for measuring and predicting the ERV of a fragment in a particular language, checking the regression equations by proportional analysis of the ERV values obtained using the equation and using a non-mediated analysis of the arithmetic mean values of the respondents' answers, comparing the ERV values of the same fragments in different languages, confirming or refuting the advanced research hypotheses, ERV parameters are ranked according to the values of standardized Beta coefficients (see the Coefficients table "below).

The proposed method is illustrated by screenshots from the site where the online application is located:

FIG. 1 Main page of the online application form;

FIG. 2 Web page of an online questionnaire with one of the fragments;

FIG. 3 Screenshot of a fragment of an online questionnaire in Hebrew;

FIG. 4 Block diagram of a device for implementing the method.

The method is as follows.

To increase the objectivity of a random sample, choose the type of discourse for analysis and select material (books, speeches, speeches, etc.) with a volume of at least 40,000 words within this type of discourse in one of the languages studied. Fragments of discourse are selected by random sampling, since any of them has some kind of real-life discourse. RV cannot be zero, it can only tend to zero. From each fragment, parameters associated with PB are distinguished. Of all the parameters, those that are able to act on the recipient are selected. These are the parameters due to which the elements of the discourse fragment (words, phrases, etc.) move from a neutral style to another (official business, conversational, jargon, etc.), namely: metaphor, comparison, repetition, stylistic contrast, non-stylistic contrast, epithet, irony / satire, rhetorical question, exaggeration / understatement, pun (wordplay), neologism, allusion, rearrangement of parts of a word / sentence, replacing a word / structure with another word / structure, personification, zevgma, gradation. This is the so-called rhetorical techniques, the purpose of which is to increase the ERV of the fragment. Separately, the density parameter of rhetorical devices is distinguished.

After selecting the parameters, large sections of the discourse are distinguished from the selected material in one of the languages studied.

They compile a standard list of ELFs selected from large sections of the discourse: ELFs of space, time, social ELFs (belonging to the social class, social needs, the number of speech recipients, social tasks of the historical period, the activity of public life and the degree of participation in it), actually political ELFs ( the intensity of communication with the history of society, political goals and objectives of the historical period), psychological ELF (personal relations between subjects of discourse, mental state).

On the basis of the standard ELF list, ELFs are distinguished that affect specifically the ERW in the studied languages by proportional analysis of the parameters in the discourse fragments; get a specialized ELF list.

They compose a table where they indicate the proportions of one parameter to the total number of parameters in a particular fragment of the discourse:

In this way, the relative magnitude of the presence of a parameter in a discourse fragment can be determined. Correlation of selected parameters with discourse fragments allows ranking the latter according to any of these parameters. Such a ranking in the distribution of each ELF, for example, in decreasing order, allows for a comparative analysis of the ranking of fragments of discourse and distribution of ELF to distinguish between them.

Example: according to Tab. 1 distribute fragments of discourse in descending order of ELF space, i.e. to reduce the volume of space (stadium (fragment A) - area (fragment B) - official tribune (fragment C). Add a column with the specified ELF and compare the values for a stable relationship (correlation) between the specified ELF and the parameters:

A stable dependence refers to such a ranking of discourse fragments according to one of the parameters that completely correlates with the distribution of ELF, and none of the values of the parameter or distribution of ELF violates the general tendency.

Stable relationships between ELF and parameters are obtained. If stable relationships are established between a certain ELF and at least one of the parameters, then the ELF data are referred to as ELFs affecting the ERV. In the case of Table 2, there are stable relationships between the amount of discourse space and parameters 2, 14, 15 (directly proportional); 5, 13 (inversely proportional).

Using a similar procedure, all ELFs from the standard list and all parameters are checked for stable dependencies between them. Make a specialized list of ELF (those that affect the ERV).

In order to conduct a psycholinguistic study of the ERV, which is as free as possible from interference of extralinguistic factors, the influence of ELF from a specialized list is leveled. In other words, when conducting the study, fragments with the same ELF are left from the specialized list. For example, all fragments of the discourse were pronounced in a television studio.

The researcher may introduce additional extralinguistic selection criteria, if their effect on the ERV is obvious. These criteria include, for example, authorship.

Thus, a set of discourse fragments with the same ELF is obtained: all fragments are in equal extra-linguistic conditions.

Then, fragments with 18 parameters selected in accordance with the number of parameters — 18 fragments are selected from this set of fragments, 2 fragments are needed to compare the densities of rhetorical devices, 2 fragments in which there are no rhetorical devices, and 1 control fragment. Total 23 fragments. The control fragment is selected so that it contains 100% logical, encyclopedic and accessible to all information. The control fragment identifies those respondents who considered it illogical and unconvincing, which means that they did not go into the essence of the fragment, but assessed their opinions and feelings about how convincing this fragment sounds. This is one of the objectives of the study - to collect the answers of respondents who did not go into the essence of the fragment and were guided not by logic, by their answers.

knowledge, and sensations, psychological reaction and quickly formed opinion about the contents of the fragment.

To evaluate the parameter “density of rhetorical devices”, it is necessary to introduce 3 fragments with the most different densities: low, medium and high: 0.005; 0.01; 0.02. The indicated densities can vary from language to language from one array of fragments to another. Thus, a set of original discourse fragments is obtained in which the selected parameters are mixed and the latter are contained in various quantities. To increase objectivity, studies equalize the volume of fragments and the number of parameters. The volume of the fragment equal to 400 printed characters is a sufficient amount for the implementation of any rhetorical device. The number of parameters is equalized by leaving 1 parameter in each fragment. As a result, a set of 18 modified fragments containing 18 selected parameters is obtained, as well as 2 additional fragments for comparing contrast densities in the correlation analysis, 2 fragments without parameters as a reference standard for RV, and 1 control fragment. Fragments with a standard of standard RV are introduced, firstly, in order to fully verify that the presence of parameters of the RV leads to a deviation from the norm and a subsequent increase in the ERV, and secondly, to determine how much the ERV of fragments without selected parameters differs from fragments with them .

Then, from the fragments obtained, an online questionnaire is compiled. This process includes both the preparation of the questionnaire and the development of special software for posting it on the Internet. To obtain a good percentage of completed questionnaires, divide the set of fragments into 3 approximately equal parts and get 7 + 7 + 6 (20) fragments in online questionnaires and add 2 fragments to each of them without rhetorical techniques (i.e. without parameters) and 1 control fragment. Accordingly, they come to the following: 10 + 10 + 9 fragments in 3 online profiles.

All fragments in online profiles are organized according to the following two principles. In the main part (i.e., in the middle of the questionnaire from the 3rd to the 7th fragments), the most difficult to understand fragments are located. Difficult to perceive fragments alternate with simpler ones.

The period for conducting an online questionnaire is set within the shortest possible time — 3 months, since ELFs may appear over longer periods that are independent of the researcher or the conditions under which the study is conducted.

Set the time range during which the online application should be filled: the maximum range is from -60% (in min.) To + 50% (in min.) Of the total number of fragments in this online application. Accordingly, for an online questionnaire in 9-10 fragments, this range is 4-15 minutes. The time range limitation is introduced both to avoid the quick automatic filling of the online questionnaire and to avoid questionnaires, where

the respondents tried too deeply to understand the essence of the fragment, which does not correspond to the objectives of the study and the instructions for the online questionnaire that the respondents read before filling it out. An online questionnaire checks and records the credibility of fragments - their ERV - and not their logic, morality, etc.

The fragments use grammatical constructions of no higher than average complexity, a small amount of specific terminology of the studied type of discourse can be used. Passive constructions, jargon, abbreviations, and complex / rare words are avoided in fragments to reduce perceptual complexity.

Each online questionnaire consists of an introductory, requisite classification and information part and a second thanks.

- The introductory part contains general information about the research institute, the purpose of the study, the practical application of the research results, as well as information about the usefulness of this questionnaire, which increases the motivation of respondents.

- The requisite part provides information about the institution conducting the survey, the name of the questionnaire, time, topic.

- The classification part (passport) requests information about the respondent: gender, age, education and email address (optional for those who would like to receive the results of the study to their email address)

- The information part contains introductory instructions for completing, 6 basic questions-fragments (4 with parameters, 2 without and 1 control fragment)

- Repeated gratitude for passing the questionnaire is at its end.

Any elements in fragments that could incline the respondent to one or another variant of the answer were eliminated as much as possible. Such elements are widely known under the name of effects (in foreign literature - biases). Below is a table with all the possible effects in the online questionnaire and ways to neutralize them, which are used in the preparation of the online questionnaire for accurate measurement of the ERV:

As a response option, a balanced 5-point scale is used: 2 positive answers (“convincingly” and “absolutely convincing”), 2 negative (“unconvincing”, “completely unconvincing”) and 1 neutral (“difficult to answer”). The presence of discussion topics in fragments made all possible answers equally possible, which is extremely important for obtaining objective results.

In addition, fragments in each of the 3 online profiles are not related either thematically or logically. The questionnaires take into account the specifics of culture and use realities that are understandable to most respondents who speak a particular language. The online application is widely used illustrations. When filling out an online application, complete anonymity is provided. The computer program for online questionnaires is designed for the purpose and objectives of the study.

It is written in HTML and Javascript using PHP to process data on the server. The jQuery library is also used to dynamically display content: loading new fragments of the questionnaire without reloading the page. Data on the responses of the respondents is recorded in the MySQL database for further work with the obtained data set.

New technical solutions undertaken in the development of software for online profiles.

1. Questionnaires alternate with each new visit to the site in cycles: questionnaire No. 1, 2, 3. As a result, the number of respondents who answered each of the 3 online questionnaires will be approximately the same.

To do this, in the choose_tbl table in the SQL database, there is a column called 'number', which takes values 1-3. After the respondent clicks “Go to the questionnaire”, the program contacts the database to select the next questionnaire.

After that, the program updates the value in the column 'number' so that the next respondent is shown a different questionnaire. There are 3 online profiles in total.

The program code looks like this:

2. A solution was found for displaying text in languages that are graphically very different from European ones: Hebrew, Chinese, Japanese, etc.

Imagine a simple record of a part of the program in Russian:

«Нельзя просто сидеть в парламентах и обслуживать нынешнюю власть. Я объявляю войну вранью, безответственному популизму и пустым, невыполнимым обязательствам. Я первый независимый политик, и могу говорить о том, что происходит на самом деле...Нам необходимо смотреть проблемам в лицо и взять на себя груз, оставшийся после очень трудного периода истории страны.

“You can’t just sit in the parliaments and serve the current government. I declare war on lies, irresponsible populism and empty, impossible obligations. I am the first independent politician, and I can talk about what is really happening ... We need to face the problems and take on the burden left after a very difficult period in the history of the country.

Our country is a country for great things. This is a country of unlimited possibilities. ”

In Hebrew, in contrast to the Russian language, the direction of the text is opposite, i.e. from right to left. Most browsers that display web pages and are accessible to ordinary users are developed in America or Europe: Internet Explorer (USA), Google Chrome (USA), Safari (USA), Opera (Norway) - therefore, they are mainly designed for European languages. Therefore, when trying to simply write a paragraph in an Hebrew HTML document, the words in sentences are rearranged:

Any character — in our case, a language letter — can be encoded, and if you encode each letter with a number, the string of numbers is written in the HTML document in the direction that the browser reads correctly, i.e. from left to right. Therefore, the record was presented in Unicode encoding:

The result obtained is shown in FIG. 3.

Moreover, this form of writing solves the problem of multidirectional text: for example, when one part of a sentence is written in a language where the direction of the text is from right to left

(Semitic-Hamitic language family), and the other part is in a language where the direction of the text is from left to right (Slavic, Romano-Germanic group), or from top to bottom (Altai language family (Japanese, Korean, etc.) and Sino-Tibetan family (Chinese and etc.).

3. To exclude unfilled questionnaires (thereby improving the quality of all received data), as well as those online questionnaires that were filled in for a period of at least 4 minutes and no more than 15 minutes, a responder logged in and out of the site was entered into the program. The entry time to the site is recorded at the moment you click the "Go to profile" button. The program inserts into the database table corresponding to this questionnaire (survey_1, survey_2, survey_3) a new row with the selected parameters (date and time of the beginning of the passage and completion of the online questionnaire, language).

The respondent’s new serial number (his unique identification number), the time and date of the start of filling out the questionnaire, and the language in which the questionnaire was filled out are recorded in the SQL database. Further, after answering the last question, the time and date of completion of filling out the questionnaire is recorded. Thus, the researcher receives a set of respondents who fill out the questionnaire completely.

We get the filling start date using the built-in MySQL ′ NOW () ′ function.

The program code is shown below:

4. Each fragment to the respondent is shown on a separate web page. “As if,” because the change of fragments is done using JavaScript (jQuery library) without reloading a new web page. Those. in reality in program code

all fragments are recorded in one file, but when displayed in the browser, part is hidden for the respondent, so he sees the web page partially - at a time only one fragment is displayed for him. Accordingly, the browser uploads to the web page all 8-9 fragments at a time, but the fragments are displayed in turn, and only when the current fragment is answered the next one will be shown and the current one will be hidden. The program code looks like this:

To obtain objective results and accurate statistics, you need at least 200-300 answers for each of the parameters and in each language studied. In our example, with 18 different parameters and 2 additional fragments, these are 900 (300 * 3) completed online profiles in each language studied.

So, 3 online questionnaires have been compiled for assessing, measuring, comparing and predicting ERV. Example fragments from one online application:

1. (without parameter): “I would like to say that elections are one of the manifestations of democracy, and personal participation in them can decide what the new parliament will be like. Make an informed decision and come to the polls. ”

2. (repeat): “When you say today that you have built factories. They were built by prisoners and soldiers of the Soviet army. 75% off! And Belomorkanal - only prisoners. Let’s put two million in jail and throw in the construction of factories, they will build thousands of factories, thousands of factories! ”

3. (comparison): “Our country should not compete for the state budget, which is limited and small, but for the funds that private and international business has. Therefore, we will continue to consider these issues: the judicial system, the competitive environment and private business as an engine of change ”;

4. (rhetorical question): “Based on what the previous speaker said, a hundred years ago it was also: everyone spoke. Ok, we went on the left path. At night in

On October 17th, the left took power. They walked 75 years. Well, where is the result? Moreover, this power had everything in its hands ”;

5. (stylistic contrast): “I participated in the presidential election and wrote more than once. And now the new president, the whole country rebooted, and everyone thought what the new president would be, he has such an Iphone, he’ll now tweet the Internet somewhere, google it, it will solve everything. And nothing has changed ”;

6. (control fragment): “We need to strengthen the role of state institutions in society. Sustainable development of society is impossible without a capable state. A strong state is a state whose goal is to create the best, most comfortable conditions for life, creativity and entrepreneurship ”;

7. (metaphor): “I am a doctor of science, and I understand something in science! Now I am at the head of the party, and there is no need to teach me. Here you are on the legacy of Marx. You should be ashamed: you are eating the inheritance of a party that is no longer there. That was the party! ”;

8. (non-stylistic contrast): “We understand that there are people who have interests in having some part of the territory of our country separate. We would like to unite around ourselves, including other people for whom the integrity of the country is really an absolute priority ”;

9. (without parameter): “As for the basic specialties: teacher, doctor, engineer, scientist and military and others, the salary should be doubled. I will do everything necessary to restore relations with neighboring states. For young people I guarantee free education up to higher education. ”;

10. (epithet): “You can’t just discuss in parliaments. I declare war on lies, populism and empty obligations. I’m an independent politician, and I can talk about what really happens ... We need to solve the problems that remain after a difficult period in the history of the country. ”

All 3 online questionnaires are translated into the studied languages while maintaining the potential of the real estate. In order for the potential of the RW to be the same in fragments in different languages, the quantitative and qualitative characteristics of the discourse are preserved. Quantitative characteristics include: the frequency of use of the word, their joint occurrence and distribution in the discourse, the volume of the fragment. Qualitative characteristics include the presence of rhetorical devices. To preserve the qualitative characteristics of the fragment, rhetorical techniques with the same ERV are excluded and / or added, or replaced. For example:

“Stolypin spoke here, the centrists and the right supported, the left did not support”;

"Stolypin spoke here, the centrists and the rightists supported him, the leftists did not." (Lexical repetition replaced with an ellipsis);

As you know, a violation of a design accepted in the language is used as a stylistic device, therefore, when translating a sentence from Russian, where one technique is used, in our case, lexical repetition, into English, it turns out that the text of the translation of this sentence has two methods: lexical repetition and violation accepted in the design language.

Therefore, instead of lexical repetition, we use an ellipse. In this case, the number of uses of the ellipsis increases, compensating for the decrease in the number of lexical repetition. Ellipsis refers to the rhetorical method of replacing a structure with another structure.

Thus, when translating, the potential of the RV fragment is maximally preserved, so that the process of translation from one language to another has the least impact on the results of the study.

After the translation of the online questionnaire into the studied languages, an online survey of native speakers is carried out. All answers are automatically registered in a specialized database:

So, each parameter corresponds to 1 fragment, and for each this fragment there are, say, 300 responses from respondents who rated its ERV from 1 to 5; with 300 answers in each of the studied languages. There are also 300 answers to 2 fragments in which the selected parameters are missing.

Using the [select * from survey_1 (2, 3) timediff (end_date, start_date)> 240 and timediff (end_date, start_date) <900 operator, exclude online profiles filled out in less than 4 minutes or more than in 15 minutes. Using an additional filter [where lang = ru (eng, hebr, etc.)], online questionnaires in one of the studied languages are selected.

Then select the totality of all the answers received as a result of filtering to 3 online profiles (which were previously divided into the original 20 fragments) using the [union all] operator. Export the database in Microsoft Excel format. Combine 3 tables obtained in accordance with the answers to 3 online questionnaires, into one, where there are 18 columns with answers corresponding to 18 selected parameters, 2 columns with additional rhetorical densities (medium and high), 2 columns with respondents' answers a fragment in which the selected parameters are absent (nonparameters) and 1 column with answers to the control fragment - see Table. 5.

Оставляем один из столбцов с непараметрами вне выборки для проверки уравнения регрессии в дальнейшем. Вычисляем среднеарифметические значения (САЗ) ЭРВ каждого параметра и непараметра. Этот предварительный анализ показывает общую картину и тенденции, позволяющие выделить параметры, имеющие наибольшие ЭРВ.

We leave one of the columns with nonparameters outside the sample to check the regression equation in the future. We calculate the arithmetic mean values (SAZ) of the ERV of each parameter and nonparameter. This preliminary analysis shows the general picture and trends, allowing to identify the parameters that have the highest ERV.

However, the analysis of SAV ERV is superficial, since it does not take into account the so-called “Heavy tails”, in the presence of which the distribution of the ERV values does not obey the normal law. Therefore, the results of the analysis of SAS ERV are confirmed, corrected by correlation analysis for the Pearson coefficient, which is a multiple linear regression. In accordance with this analysis, correlations between the dependent and independent variables are found, where the ERV of the fragment acts as the dependent variable, and 23 parameters and nonparameter are independent. The values of the dependent variable are determined by the respondents; the values of the independent variables are evaluated by the researcher, transforming the qualitative parameters (metaphor and other rhetorical techniques) into quantitative (0.1). In fact, 20 two-digit dummy variables with values of 0 for non-parameters and 1 for parameters are introduced. The dummy variable representing the density of rhetorical devices is ambiguous and takes the following values: 0 - if the density parameter is not in the fragment, 1 - if the density is low (in our example - 0.005), 2 - if the density is average (in the example - 0.01) , 3 - if the density is high (in the example, more than 0.02).

All the given values of the dependent and independent variable are written in the table in the form suitable for automatically calculating the Pearson correlation coefficient and compiling the multiple regression equation with these variables:

Parameter and non-parameter numbers see Table 5.

All calculations are performed using the Microsoft Excel, STATISTICA, UNISTAT 6.5, MedCalc, SPSS 20, SAS software packages at a significance level of p <0.05 (95% probability). In other words, all the results obtained with a probability of 95% will be true for other representative samples within the studied languages and types of discourse (if initially fragments were selected from only one type of discourse).

Using the correlation analysis, the following data are obtained: Pearson's correlation coefficient for the entire regression model, for partial correlations between the ERA and parameters / non-parameters, and for partial correlations between parameters / non-parameters; variables included in the analysis (since the analysis includes variables whose values have a normal distribution); the coefficient of determination for the entire model (this coefficient determines the quality of the model) and its change (determines the reliability of the analysis results); F change (also determines the reliability of the analysis results); F-ratio (shows how random the relationship between the variables is); non-standard coefficients B for the regression equation; standardized Beta coefficients (determine the contribution of each parameter to the ERV value); collinearity diagnostics (checking the model for collinearity is carried out to identify a close correlation between the parameters selected for analysis, which jointly affect the ERV value, which makes it difficult to objectively evaluate each parameter individually).

For the data in Table 7, the results of correlation analysis are as follows:

Descriptive statistics

Analyze the data.

Using the correlation coefficient R for the entire model, it is determined whether there is a relationship between the ERA and the parameters and how strong it is. So, in our example, the highest R in model 3 is R = 0.476, which indicates the existence of a weak correlation between the ERV and the variables 2, 10, 13 included in the analysis, namely, rhetorical methods of comparison, puns and permutation of parts of a word / sentence. Model 3 includes these three parameters.

The Pearson coefficient for particular correlations between the EDS and a specific parameter / non-parameter allows us to determine whether there is a connection between them, the direction of this connection, and the strength. In our example, no such correlations were revealed.

The determination coefficient R ² shows the quality of the entire regression model: the larger R ² , the higher the quality of the model. So, in our example, model 3 is the highest quality, since R ² = 0.227.

A change in R ² (R ² change) occurs as a result of adding / removing a parameter from the regression equation. Accordingly, if the magnitude of the change in R ² is large enough, then this model is a good predictor. In our example, this is model 3, R ² change = 0.084.

F Change, that is, the ratio of the proportion of variation of the dependent variable (ERV) to its fraction of variation attributable to the residuals, demonstrates that the Pearson coefficient R for model 3 with 3 parameters is more reliable than for model 1 with the parameter “comparison”, or model 2 with the parameters "pun" and "permutation of parts of the word / sentence."

Using the F-relation, or the so-called F-statistics determine whether the established correlation between variables is random or non-random. The F-ratio for all 3 models is insignificant, which does not allow us to confidently reject the null hypothesis that the connection between the parameters “comparison”, “pun” and “rearrangement of parts of a word / sentence” with EER is random. Accordingly, the relationship between these parameters and the EDS may be random.

In accordance with non-standardized coefficients, they compose a regression equation, which, when substituting parameter / non-parameter values (in our case, these values are 0.1 or 0-4 (the density of rhetorical devices)), allows to measure the ERV of a particular fragment, as well as to predict the ERV for, for example, modeling an ideal fragment from the point of view of ERV:

ERV _(fr.) = 1,574a + 1,574b + 1,574s + 3,426,

where a is the value of the parameter “comparison”, b is the value of the parameter “pun”, c is the value of the parameter “permutation of parts of the word / sentence”.

In predicting and modeling, it is important to know which parameters are fundamental and have the greatest influence on the magnitude of the ERV. To do this, calculate the standardized coefficients of the regression equation. So, in our example, all 3 parameters make the same contribution to the ERV value (0.29).

Diagnostics of collinearity reveal the presence of relationships between variables, i.e. parameters / nonparameters. Such relationships can affect the results of the entire correlation analysis, therefore it is important to diagnose the absence of multicollinearity in the regression model. In our example, collinearity indicators have the following meanings: Eigenvalue (eigenvalue of collinearity) = 1,378; one; one; 0.622 (less than the critical value of 13) and Condition index (conditionality indicator) = 1; 1.174; 1.174; 1,488 (less than the critical value of 15). Therefore, the absence of multicollinearity in the regression model 3 was noted.

Verification of the regression equation is carried out using real sets of respondents' answers - a proportional analysis, which compares the ratios of the arithmetic mean values of the ERV (including fragment 22, which did not participate in

correlation analysis) and the ERV values of the same fragments obtained using the regression equation. If the difference between the obtained proportions is greater than 0.2, then this equation is not considered accurate. According to the regression equation:

1. Fragment No. 1 of questionnaire 1 (1 parameter): ERV = 1.574 * 0 + 1.574 * 0 + 1.574 * 0 + 3.426 = 3.426

2. Fragment No. 4 of questionnaire 2 (10 parameter): ERV = 1.574 * 0 + 1.574 * 1 + 1.574 * 0 + 3.426 = 5

3. Fragment No. 3 of questionnaire 3 (15 parameters): ERV = 1.574 * 0 + 1.574 * 0 + 1.574 * 0 + 3.426 = 3.426

4. Fragment No. 10 of questionnaire 3 (22 parameters): 1.574 * 0 + 1.574 * 0 + 1.574 * 0 + 3.426 = 3.426

According to the SAZ ERV:

1. Fragment No. 1 of questionnaire 1: ERV = 8: 3 = 2.667, where 8 is the total ERV value, 3 is the number of online profiles

2. Fragment No. 4 of questionnaire 2: ERV = 15: 3 = 5

3. Fragment No. 3 of questionnaire 3: ERV = 8: 3 = 2,667

4. Fragment No. 10 of questionnaire 3: ERV = 9: 3 = 3

The resulting proportions:

1.3.426: 2.667 = 1.3

2.5: 5 = 1

3.4.426: 2.667 = 1.3

4.3.426: 3 = 1.1

Thus, it is established that the relationship between the ERV and the parameters “comparison”, “pun” and “rearrangement of parts of the word / sentence” exists and is directly proportional, which confirms the initial hypothesis that fragments of discourse with rhetorical devices have a higher ERV, than fragments without them, in terms of comparison, pun and permutation of parts of a word / sentence.

However, the large difference between the proportions of the ERV values obtained using the correlation analysis and questionnaire indicates the approximation of the equation. Therefore, we do not consider the addition of the remaining parameters to the equation to be critical, especially since the hypothesis was confirmed for some of the parameters and it is less likely that fragment No. 10 of questionnaire 3 (parameter 22) has the same ERV as fragment No. 1 of questionnaire 1 (1 parameter ) and fragment No. 3 of questionnaire 3 (parameter 15), although according to the equation this is so.

In this regard, the remaining parameters are introduced into the regression equation with non-standardized coefficients calculated for them, which were previously excluded by step selection, since their distribution did not correspond to the normal one.

ERV _(fr.) = 0.148a + 1.574b + 0.177s + 0.112d + 0.018e + 0.177f + 0.083g + 0.018h + 0.177i + 1.574j + 0.112k + 0.083l + 1.574m + 0.112n + 0.148o + 0.083p + 0.083q + 0.171r + 3.426, where the letters,

indicate the value of the parameter, all letters a-r correspond to 1-18 parameters in ascending order.

So, in order to measure the ERV of a fragment, it is necessary to substitute the values in the dummy variables of the equation. If the constant in the equation is much larger than the total sum of the variables, then the selected parameters are only a small part of those factors that affect the ERV of the fragment. The smaller the constant in the equation, the greater the objectivity and accuracy of the equation itself, because this means that all the main influence factors are present in the equation.

In our example, in the full regression equation, the constant is 3.426, and the maximum sum of all values is 6.406. We believe that this constant value is large (more than 0.5 of the maximum value of the aggregate of all parameters of the electric propulsion) and that in the initial hypothesis some significant influence factors remained unaccounted for.

In this case, basically, we can state the greater reliability of the compiled regression equations, since there is no multicollinearity in them.

Compare the results of the analysis of SAZ ERV with the results of the correlation analysis. The presence of fully and partially confirmed hypotheses and trends is determined. In this case, a hypothesis is understood as an assumption that, for example, a fragment with a rhetorical method “metaphor” has a larger ERV than without it. Hence the values of the dummy variable (ERV): with a metaphor - 1, without - 0.

The gradation of hypotheses to fully and partially confirmed and trends is based on a combination of data from the analysis of arithmetic mean values and correlation analysis:

we refer to fully confirmed hypotheses, which were confirmed during the analysis of the arithmetic mean values of the electric propulsion and have a correlation coefficient of at least 0.1, provided that this coefficient is reliable;

partially confirmed - hypotheses that were confirmed in the analysis of the arithmetic mean values of the electric propulsion and having a correlation coefficient less than 0.1, but more than 0.05, provided that this coefficient is reliable; if the coefficient is not reliable, then those hypotheses that have a correlation coefficient of at least 0.1;

to trends - hypotheses, either confirmed during the analysis of arithmetic mean values of the electric propulsion and having a correlation coefficient less than 0.1, or having a correlation coefficient not less than 0.1.

According to the indicated gradation, we identify in the example:

- fully confirmed hypotheses: fragments of discourse containing rhetorical methods of comparison, and / or pun, and / or rearrangement of parts of a word / sentence, and / or

medium or high density rhetorical devices have a significantly higher ERV than fragments containing other rhetorical devices, and, especially, without them;

- partially confirmed hypotheses (in accordance with the partial correlations in the Excluded Variables table, the Partial correlation column): parameters 3, 4, 6, 9, 11, 14 have ERS EAS above 3.0 and the value of partial correlation above 0.1, t. e. fragments of a discourse containing rhetorical techniques of repetition, stylistic contrast, epithet, exaggeration / understatement, neologism and replacing a word / structure with another word / structure have a higher ERV than fragments containing other rhetorical techniques (except those indicated in fully confirmed hypotheses), and, moreover, without them;

- trends: in relation to parameters 5, 7, 8, the initial hypothesis was refuted and the opposite was put forward, since the partial correlation coefficients have a negative sign, which means that the correlation is inverse. In other words, in the language under study there is a tendency for the ERV fragments to decrease if the fragment contains parameters 5 (non-stylistic contrast), 7 (irony / satire) and 8 (rhetorical question) .. It should be noted that the whole analysis was carried out on a representative example of 3 respondents , therefore, in a full statistically significant sample, the results of the study will be different.

The reliability of partially confirmed hypotheses and trends can only be verified by additional research that takes into account the factors that cause the exclusion of variables from the regression equation.

After confirmation / refutation (including partial) of the hypotheses put forward, the results of the analysis of the SAZ ERV and the correlation analysis for all the languages studied are compared and presented in tabular form for the convenience of comprehension of compressed information:

Make a list of common and unique for the studied languages traits and patterns:

Общее (для 3 языков):

General (for 3 languages):

the more “repeat” in the fragment of the rhetorical device (parameter 3), the higher the ERV of the fragment;

the more “stylistic contrast” in the fragment of the rhetorical device (parameter 4), the higher the fragment's ERV;

the more “non-stylistic contrast” in the fragment of the rhetorical device (parameter 5), the higher the ERV of the fragment;

the more “neologism” in the fragment of the rhetorical device (parameter 11), the higher the ERV of the fragment;

the higher the density of rhetorical devices (parameter 18) in the fragment, the higher its ERV.

Общее (английский + русский): и т.д. Данный способ может быть реализован с помощью программно-аппаратного комплекса (ПАК), включающего в себя блок обработки данных, базу данных, блоки ввода и вывода (фиг. 4).

General (English + Russian): etc. This method can be implemented using a hardware-software complex (PAC), which includes a data processing unit, a database, input and output blocks (Fig. 4).

The input unit and the output unit interact with the PAC respondents and the operator. The input block (online questionnaire) is used to enter the responses of respondents. The output block (MySQL Workbench 5.2 CE program) carries out data output in a format convenient for subsequent processing by the operator (in table formats CSV, Excel Spreadsheet, etc.).

The PAK database (DB) contains information about the selected parameters and their values, information about respondents, etc. The data processing unit interacts with the database of the hardware and software complex, input and output blocks that perform software processing of information from the database, and stores the processing results in the database.

As a PAC, a personal computer with the Windows 7 operating system can be used, containing a central processor, random access memory, a hard disk drive, a keyboard, a monitor, a network card, and / or a model for providing work with an Internet resource. The PAK operation algorithm is shown in FIG. 1. The respondent evaluates each of the parameters according to the selected criterion - in accordance with the proposed scale of 5 answer options. This assessment is sent to the database of the hardware-software complex, which has an array of information for evaluating the parameter according to the selected criterion - the effectiveness of speech exposure - in a 5-point system. At the same time, an array of information is processed in accordance with the types of downloaded information: information about the respondent, online questionnaire and respondent's answers, each of which the PAC classifies into information elements recorded in separate database cells. For example, information about the respondent includes the following information elements: respondent's name, last name, email address, age, gender, education. Each response of the respondent is assigned a corresponding value according to a 5-point system. For example, the PAK attributes the value “1” to the answer “completely unconvincing”, and the value “5” ascribes “absolutely convincing”. Thus, we obtain the value of the effectiveness of speech exposure for each fragment separately, as well as for each parameter as a whole, after sorting the database using the function "select concat (name of the information element) from" name of the online questionnaire "where lang = ′ name of the language online profiles' ”in the output block (MySQL Workbench 5.2 CE program).

At the next stage of the method’s work, the information available in the database is extracted by pressing the “Export” button by the PAC operator and exporting this information in CSV, Excel Spreadsheet, HTML, XML, and other formats suitable for the subsequent preparation of the regression equation and prediction of the effectiveness of speech exposure (ERV) ) In addition, the PAC may contain additional functions and information elements that provide ranking of fragments of the discourse according to the EDS and prediction of the ERA of these fragments.

The claimed method was tested in evaluating the ERV of 19 fragments by 450 respondents. Each fragment was evaluated according to the selected parameters and criteria - ERV (degree of persuasiveness). The resulting set of quantitative estimates for each parameter made it possible to rank the parameters and the fragments themselves by the EDS and compose a regression equation capable of predicting the ERA of any fragment by substituting the parameter values into this equation.

The advantages of the claimed method are:

1. a high degree of accuracy and reliability;

2. reliability and validity;

3. the ability to verify the main result obtained during the study - the regression equation

4. the ability to directly study the effectiveness of speech exposure in different languages and at different stages of their development;

5. the possibility of measuring and modeling the ideal from the point of view of the ERV fragment of the discourse for different purposes;

6. the ability to automate the collection, storage and analysis of research results;

7. The claimed method is an open system, can be combined with other methods (for example, with ordinary paper questionnaires) and form a hybrid method;

8. the speed of the method;

9. the possibility of remote study of the ERV and the lack of contact with the questionnaire researcher;

10. lack of sophisticated equipment to obtain the necessary information;

11. low cost (pays off if the sample exceeds 300 respondents);

12. anonymity of online profiles;

13. convenience of the respondent (filling in at any place where there is an Internet network, at any convenient time, etc.);

14. the possibility of using multimedia (increases the percentage to the end of the completed questionnaires and the motivation of respondents).

Claims (2)

1. A method for predicting the effectiveness of speech exposure (EVR) of discourse fragments in different languages, characterized in that they select the parameters, act on the recipient in one of the languages studied, which are used as rhetorical devices; model discourse fragments, each of which contains one of the selected parameters; distinguish extralinguistic factors that affect the effectiveness of the speech exposure of the fragments in the studied languages and level their influence in the modeled fragments by modifying these fragments; equalize the volumes of fragments for each parameter, compose an online questionnaire, including modified fragments with selected parameters, add a control fragment; translate the online questionnaire into the studied languages while maintaining the potential for speech exposure; conduct online surveys of native speakers; conduct a correlation analysis for the Pearson coefficient for each language separately to draw up a regression equation; carry out a comparative analysis of ERV fragments with various parameters in the studied languages; quantitatively determine the contribution of the parameters to the ERV value and rank them according to the values of standardized Beta coefficients, the conclusion about the effectiveness of the speech exposure of the fragment is made in accordance with the ERV value obtained by calculating the regression equation. 2. The method according to p. 1, characterized in that rhetorical techniques are chosen as parameters, such as: metaphor, comparison, repetition, stylistic contrast, non-stylistic contrast, epithet, irony / satire, rhetorical question, exaggeration / understatement, pun, neologism , allusion, permutation of parts of a word / sentence, replacement of a word / structure with another word / structure, personification, zevgma, gradation, permutation of parts of a word / sentence, density of rhetorical devices.

Images