Why we put content first
Any user - and this is an active or potential student - interacts not with the “bare” code, neural networks and methods, but with content. And the interest in learning directly depends on the quality of this content.
Today in the blog we will share thoughts about why content in e-learning in general and in Puzzle English in particular is so important and the technologies with which we deliver it are not so important.
Let's start with the most important thing - the user should understand both the content itself and the content distribution system on the site. This means that it is not enough to write down the task in detail or explain in detail the rule for the use of articles, it is also necessary to submit all this competently.
')
Therefore, the learning process with us is based on simple principles: consistency, structuredness, coherence. And in each of them another is manifested - flexibility.
Consistency is manifested in the fact that the user follows a consistent program, competent from a methodological point of view. This may be a business English course, and Puzzle Academy - for those who are preparing for independent travel abroad, and any courses on the "Tichera method." In all cases, there will be transitions from simple tasks to complex ones, setting specific goals, and training with an emphasis on topics important for a particular student. And yes, flexibility will also be in the possibility of “starting” from a certain level, skipping already well-studied topics. In general, the whole process is subject to a certain logic, a system.
The second important principle is structuredness . This is why all content is categorized:
- by topic (it is obvious that all users have their own goals and interests);
- by levels of difficulty and levels of English (from Beginner (A1-A2) to Advanced (C1-C2));
- on the linguistic aspect (we take into account the content of the language unit to be studied - adjectives, nouns, verb forms, idioms, and so on);
- by style (determining the age and interests of the student, we can predict which style of content will interest him).
Thanks to these categories, it is easier for us to offer students suitable programs, and it is easier for students to find courses that are interesting for themselves. Such a structure is most effective: if the lessons are arranged in alphabetical or even chronological order, it is very difficult to find the one you need.
Let's not hide: we have been testing different models for a long time and continue to search for the perfect solution. Earlier, for example, grammar exercises were divided only by levels and the date of addition - it was chaos, it was very difficult to find something concrete. But over time, we added division by topics and teachers, and everything became very simple and clear for users. In the near future, we will build an even clearer “hierarchy” of grammar exercises, so that all tasks go strictly from simple to complex.
Another important principle is connectivity . Language does not exist outside the context, and we strive to create it in different modules. For example, a student watches an excerpt from a film with subtitles, hears and sees an unfamiliar word, adds it to the dictionary. But this is not enough to memorize and learn it, so we offer to work it out in games or in word training. And vice versa - while studying a word in an exercise, a person can get a video with an example of usage: he will listen to how live actors and TV presenters pronounce a new word on the air, pay attention to the context and are more likely to supplement their own lexicon.
Naturally, it is not enough to create simply understandable content - the user should also be interested. Not all people who want to learn English, it is important which technologies will be used in training. The main thing is to have an effect. And that was fun, of course.
We will talk about learning effectiveness in the next paragraph, and in this we will discuss the multimodality of our system. So, we have several forms of educational content delivery that help us keep the user's attention and keep them interested.
Dictionary . These are not only foreign words with text translations, but also transcriptions, and audio, and even video with examples of pronunciation. At the same time, the examples are very different: frames from cartoons, fragments from films and TV shows, episodes of TV shows. And the text is read by men and women, British and Americans, at a fast and slow pace (more than 20 votes in total) - so that when listening, it was clear how different a word can sound. It is very important that the word is given in context: this is how it is memorized more quickly, so we implemented the idea of the video dictionary as soon as it appeared.
For examples, specially prepared recordings of live speakers and the output of five different TTS (text-to-speech system, Vocalware and others) are used. The selection is carried out taking into account the contextual meaning of the word from the common base; now there are about 430 thousand accurately labeled video sequences with translation.
For partial markings, the Stanford Log-linear Part-Of-Speech Tagger is used . Also, when searching for translations, the frequency of words is taken into account using 100,000 words list Brigham Young University ( CLAWS-7 tagging tool) and Google Translate data: they are based on the largest number of queries and are therefore the most complete.
Audio and video puzzles. In order for such a format for presenting educational material to be possible at all, we needed precise marking of very different audio and video content over time. Unfortunately, today technologies are very limited - although progress is moving by leaps and bounds, it is not keeping pace with our desires and ideas. To provide a user with a qualitatively selected fragment of speech, an accuracy of no less than 20 milliseconds is required, while the best of such systems on live material err on average by 10–15 times. Therefore, all the markup done by hand.
Synchronous podcasts. The user can simultaneously listen to English speech, see what is heard on the screen and read the translation of that fragment of speech that he sees in front of him. Here we just apply automatic markup, as the pronunciation is almost perfect and reference.
We use the technology of “forced alignment” (forced alignment) using a ready-made transcript using a forced aligner gentle (this is an implementation of the Viterbi algorithm based on the results of speech recognition by the ASR system Kaldi ). First, transcripts are processed using the Stanford Core NLP (automatic tokenization, sentence highlighting, partial marking). Parallel translation is performed using CAT (computer-aided translation) system.
A variety of content types performs a simple function - does not let the user get bored. In addition, many formats add an element of gamification and increase excitement during training, and some even allow you to learn imperceptibly, relaxing and having fun.
Let's not forget that learning is not only games and fun. Still, the user comes to us to learn the language.
We stand for an integrated approach, so we ask students to prioritize them at the start, and then form a suitable educational program.
Anyway, we cover all the main aspects in any case:
- reading - we teach users to understand the text using podcasts with simultaneous translation, "double subtitles" in video puzzles;
- listening - we train the perception and understanding of oral speech, for this we use video and audio puzzles, “Master Phrases”, dubbing of words and phrases in the dictionary, podcasts;
- the formation of the lexicon - to learn new words and to use them helps many games (“Danetka”, “Luggage of words”, “Training of words”) and personal dictionary;
- the formation of grammatical competences - here come to the aid of a system of lessons on the “Tichera method”, video lessons with exercises from “Grammar”, “Vitaminki” and “Buns”;
- writing and translation - useful skills train personal vocabulary with contextual translations, Translayty, Translator and all modules with parallel translation (synchronous podcasts, double subtitles in video and audio puzzles, line-by-line translations in songs);
- formation of a semantic field, linguistic picture of the world - here it’s just nowhere without a “live” speech in excerpts from films, TV shows, songs in the original language, contextual translations, Puzzle English Live (analysis of simple everyday situations, video with native speakers) are still useful on Youtube, podcasts.
Each of our module involves several aspects of learning: in the process of one lesson, the user can learn to better speak English, and replenish his vocabulary, and learn more about the cultural code of native speakers, and improve the skills of perception of oral speech.
It may seem that it is too difficult, but in fact it is realistic - in life we do not use one skill “in a vacuum”, we combine it with other knowledge and competencies. So we can say that we are preparing students for existence in a different language environment.
Of course, to learn effectively, the user must always be inspired and involved in the learning process. Therefore, the content Puzzle English is interactive and personalized.
A significant part of the tasks contains an element of gamification, and this one is still interactive. And this topic can be deepened: for example, automatically introduce game mechanics into educational materials. Include automatic training of words or expressions in podcasts, add tasks for identification of missing parts of speech in reading texts, and so on. Now the learning process and games are successfully combined in courses on the "Tichera method."
There is also a more explicit form of interactive - comments to phrases. Users can ask experts questions about the use of words, the use of grammatical structures or translations and quickly get answers to them. For many, this is a very convenient format for communicating with teachers: short and to the point. In this case, of course, we strive to preserve the information content of sections and pages. So the general comments that do not carry any useful information for other users, we moderate.
And then, of course, it's time to talk more about personalization. So that each user feels direct contact with the service and feels that we work with him personally, we offer a personal training plan, word training and “Ticher method” classes.
A personal learning plan is a central component of our entire personalization system. It connects the various modules and provides the user with a meaningful and effective version of the curriculum based on the goals, objectives and level of knowledge. When calculating a personal plan, we use machine learning technologies, statistical analysis of the results of assignments and game results, as well as the user's activity history - the last item is not very actively taken into account, but in the future we plan to develop in this direction and base the program, including on the history use of the service. But it is important to note: we only offer a trajectory, the student always has the opportunity to change the plan as his goals change.
Word workouts are based on forgetting curves or Ebbinghaus curves. In short: we take into account the proven fact - 60% of the information received, as a rule, is forgotten in the first hour, after 10 hours about 35% of everything studied remains, then the process of forgetting slows down, and eventually “settles” in 6 days approximately 20% of the original information. And these 20% are quite stable - they persist even after a month. Of course, we take into account the individual characteristics of the student - the speed of progress, the amount of vocabulary and the initial level of knowledge.
The “Tichera method” classes are also personalized. The user can change the trajectory of their training, depending on their needs and goals. Very often, in the process of learning a language, people understand that they are interested in some topic much more than others. This is an excellent opportunity to study it in more detail! We support such aspirations, including and excluding certain modules, courses and exercises.
But only personalization and interactive motivation is not enough. In general, inspiration for new feats is perhaps the weakest point of all e-learning services. It is not easy for those who study remotely in free mode to force themselves to study. We solve this problem with the help of mailings with reminders about tasks, thematic digests, messages about current tasks on a personal schedule. There is also a bonus system “puzzles”, which complements the rating system - users earn bonus points for completing personal tasks, and then spend them in a special store. This domestic currency can pay for the whole course, however, you will have to work hard to begin with.
Based on our experience, we confidently declare: in e-learning, the quality and usefulness of content are of paramount importance at every stage of training. Any educational service is created for students, but they are not interested in "bare" technology. Fascinating formats, bright packaging, intuitive interface and a feeling of personal contact, along with the excitement of entertaining mechanics, easily convey educational material to users.
Now you know why content needs to be given so much attention (and how this content can be). And in the following publications we will talk more about marketing in edtech. Stay tuned!
Readers of the blog give a coupon for 500 rubles for the purchase of a subscription, which includes 8 types of training and weekly newsletters about English grammar and vocabulary. On March 8, participate in the promotion to get even more pleasant bonuses.
And for unlimited and perpetual access to all features of the site there is an “All Inclusive” tariff (the discount does not apply).
Today in the blog we will share thoughts about why content in e-learning in general and in Puzzle English in particular is so important and the technologies with which we deliver it are not so important.
All clear?
Let's start with the most important thing - the user should understand both the content itself and the content distribution system on the site. This means that it is not enough to write down the task in detail or explain in detail the rule for the use of articles, it is also necessary to submit all this competently.
')
Therefore, the learning process with us is based on simple principles: consistency, structuredness, coherence. And in each of them another is manifested - flexibility.
Consistency is manifested in the fact that the user follows a consistent program, competent from a methodological point of view. This may be a business English course, and Puzzle Academy - for those who are preparing for independent travel abroad, and any courses on the "Tichera method." In all cases, there will be transitions from simple tasks to complex ones, setting specific goals, and training with an emphasis on topics important for a particular student. And yes, flexibility will also be in the possibility of “starting” from a certain level, skipping already well-studied topics. In general, the whole process is subject to a certain logic, a system.
The second important principle is structuredness . This is why all content is categorized:
- by topic (it is obvious that all users have their own goals and interests);
- by levels of difficulty and levels of English (from Beginner (A1-A2) to Advanced (C1-C2));
- on the linguistic aspect (we take into account the content of the language unit to be studied - adjectives, nouns, verb forms, idioms, and so on);
- by style (determining the age and interests of the student, we can predict which style of content will interest him).
Thanks to these categories, it is easier for us to offer students suitable programs, and it is easier for students to find courses that are interesting for themselves. Such a structure is most effective: if the lessons are arranged in alphabetical or even chronological order, it is very difficult to find the one you need.
Let's not hide: we have been testing different models for a long time and continue to search for the perfect solution. Earlier, for example, grammar exercises were divided only by levels and the date of addition - it was chaos, it was very difficult to find something concrete. But over time, we added division by topics and teachers, and everything became very simple and clear for users. In the near future, we will build an even clearer “hierarchy” of grammar exercises, so that all tasks go strictly from simple to complex.
Another important principle is connectivity . Language does not exist outside the context, and we strive to create it in different modules. For example, a student watches an excerpt from a film with subtitles, hears and sees an unfamiliar word, adds it to the dictionary. But this is not enough to memorize and learn it, so we offer to work it out in games or in word training. And vice versa - while studying a word in an exercise, a person can get a video with an example of usage: he will listen to how live actors and TV presenters pronounce a new word on the air, pay attention to the context and are more likely to supplement their own lexicon.
How to heat interest?
Naturally, it is not enough to create simply understandable content - the user should also be interested. Not all people who want to learn English, it is important which technologies will be used in training. The main thing is to have an effect. And that was fun, of course.
We will talk about learning effectiveness in the next paragraph, and in this we will discuss the multimodality of our system. So, we have several forms of educational content delivery that help us keep the user's attention and keep them interested.
Dictionary . These are not only foreign words with text translations, but also transcriptions, and audio, and even video with examples of pronunciation. At the same time, the examples are very different: frames from cartoons, fragments from films and TV shows, episodes of TV shows. And the text is read by men and women, British and Americans, at a fast and slow pace (more than 20 votes in total) - so that when listening, it was clear how different a word can sound. It is very important that the word is given in context: this is how it is memorized more quickly, so we implemented the idea of the video dictionary as soon as it appeared.
For examples, specially prepared recordings of live speakers and the output of five different TTS (text-to-speech system, Vocalware and others) are used. The selection is carried out taking into account the contextual meaning of the word from the common base; now there are about 430 thousand accurately labeled video sequences with translation.
For partial markings, the Stanford Log-linear Part-Of-Speech Tagger is used . Also, when searching for translations, the frequency of words is taken into account using 100,000 words list Brigham Young University ( CLAWS-7 tagging tool) and Google Translate data: they are based on the largest number of queries and are therefore the most complete.
Audio and video puzzles. In order for such a format for presenting educational material to be possible at all, we needed precise marking of very different audio and video content over time. Unfortunately, today technologies are very limited - although progress is moving by leaps and bounds, it is not keeping pace with our desires and ideas. To provide a user with a qualitatively selected fragment of speech, an accuracy of no less than 20 milliseconds is required, while the best of such systems on live material err on average by 10–15 times. Therefore, all the markup done by hand.
Synchronous podcasts. The user can simultaneously listen to English speech, see what is heard on the screen and read the translation of that fragment of speech that he sees in front of him. Here we just apply automatic markup, as the pronunciation is almost perfect and reference.
We use the technology of “forced alignment” (forced alignment) using a ready-made transcript using a forced aligner gentle (this is an implementation of the Viterbi algorithm based on the results of speech recognition by the ASR system Kaldi ). First, transcripts are processed using the Stanford Core NLP (automatic tokenization, sentence highlighting, partial marking). Parallel translation is performed using CAT (computer-aided translation) system.
A variety of content types performs a simple function - does not let the user get bored. In addition, many formats add an element of gamification and increase excitement during training, and some even allow you to learn imperceptibly, relaxing and having fun.
What is there about efficiency?
Let's not forget that learning is not only games and fun. Still, the user comes to us to learn the language.
We stand for an integrated approach, so we ask students to prioritize them at the start, and then form a suitable educational program.
Anyway, we cover all the main aspects in any case:
- reading - we teach users to understand the text using podcasts with simultaneous translation, "double subtitles" in video puzzles;
- listening - we train the perception and understanding of oral speech, for this we use video and audio puzzles, “Master Phrases”, dubbing of words and phrases in the dictionary, podcasts;
- the formation of the lexicon - to learn new words and to use them helps many games (“Danetka”, “Luggage of words”, “Training of words”) and personal dictionary;
- the formation of grammatical competences - here come to the aid of a system of lessons on the “Tichera method”, video lessons with exercises from “Grammar”, “Vitaminki” and “Buns”;
- writing and translation - useful skills train personal vocabulary with contextual translations, Translayty, Translator and all modules with parallel translation (synchronous podcasts, double subtitles in video and audio puzzles, line-by-line translations in songs);
- formation of a semantic field, linguistic picture of the world - here it’s just nowhere without a “live” speech in excerpts from films, TV shows, songs in the original language, contextual translations, Puzzle English Live (analysis of simple everyday situations, video with native speakers) are still useful on Youtube, podcasts.
Each of our module involves several aspects of learning: in the process of one lesson, the user can learn to better speak English, and replenish his vocabulary, and learn more about the cultural code of native speakers, and improve the skills of perception of oral speech.
It may seem that it is too difficult, but in fact it is realistic - in life we do not use one skill “in a vacuum”, we combine it with other knowledge and competencies. So we can say that we are preparing students for existence in a different language environment.
Well, what about motivation?
Of course, to learn effectively, the user must always be inspired and involved in the learning process. Therefore, the content Puzzle English is interactive and personalized.
A significant part of the tasks contains an element of gamification, and this one is still interactive. And this topic can be deepened: for example, automatically introduce game mechanics into educational materials. Include automatic training of words or expressions in podcasts, add tasks for identification of missing parts of speech in reading texts, and so on. Now the learning process and games are successfully combined in courses on the "Tichera method."
There is also a more explicit form of interactive - comments to phrases. Users can ask experts questions about the use of words, the use of grammatical structures or translations and quickly get answers to them. For many, this is a very convenient format for communicating with teachers: short and to the point. In this case, of course, we strive to preserve the information content of sections and pages. So the general comments that do not carry any useful information for other users, we moderate.
And then, of course, it's time to talk more about personalization. So that each user feels direct contact with the service and feels that we work with him personally, we offer a personal training plan, word training and “Ticher method” classes.
A personal learning plan is a central component of our entire personalization system. It connects the various modules and provides the user with a meaningful and effective version of the curriculum based on the goals, objectives and level of knowledge. When calculating a personal plan, we use machine learning technologies, statistical analysis of the results of assignments and game results, as well as the user's activity history - the last item is not very actively taken into account, but in the future we plan to develop in this direction and base the program, including on the history use of the service. But it is important to note: we only offer a trajectory, the student always has the opportunity to change the plan as his goals change.
Word workouts are based on forgetting curves or Ebbinghaus curves. In short: we take into account the proven fact - 60% of the information received, as a rule, is forgotten in the first hour, after 10 hours about 35% of everything studied remains, then the process of forgetting slows down, and eventually “settles” in 6 days approximately 20% of the original information. And these 20% are quite stable - they persist even after a month. Of course, we take into account the individual characteristics of the student - the speed of progress, the amount of vocabulary and the initial level of knowledge.
The “Tichera method” classes are also personalized. The user can change the trajectory of their training, depending on their needs and goals. Very often, in the process of learning a language, people understand that they are interested in some topic much more than others. This is an excellent opportunity to study it in more detail! We support such aspirations, including and excluding certain modules, courses and exercises.
But only personalization and interactive motivation is not enough. In general, inspiration for new feats is perhaps the weakest point of all e-learning services. It is not easy for those who study remotely in free mode to force themselves to study. We solve this problem with the help of mailings with reminders about tasks, thematic digests, messages about current tasks on a personal schedule. There is also a bonus system “puzzles”, which complements the rating system - users earn bonus points for completing personal tasks, and then spend them in a special store. This domestic currency can pay for the whole course, however, you will have to work hard to begin with.
findings
Based on our experience, we confidently declare: in e-learning, the quality and usefulness of content are of paramount importance at every stage of training. Any educational service is created for students, but they are not interested in "bare" technology. Fascinating formats, bright packaging, intuitive interface and a feeling of personal contact, along with the excitement of entertaining mechanics, easily convey educational material to users.
Now you know why content needs to be given so much attention (and how this content can be). And in the following publications we will talk more about marketing in edtech. Stay tuned!
If you want to upgrade English, come to us
Readers of the blog give a coupon for 500 rubles for the purchase of a subscription, which includes 8 types of training and weekly newsletters about English grammar and vocabulary. On March 8, participate in the promotion to get even more pleasant bonuses.
And for unlimited and perpetual access to all features of the site there is an “All Inclusive” tariff (the discount does not apply).
Source: https://habr.com/ru/post/350558/
All Articles