<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by OwnSlides on Medium]]></title>
        <description><![CDATA[Stories by OwnSlides on Medium]]></description>
        <link>https://medium.com/@ownslides?source=rss-419a7c8bdef8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*j8WmE_kEQLW_IplFtBvr7w.png</url>
            <title>Stories by OwnSlides on Medium</title>
            <link>https://medium.com/@ownslides?source=rss-419a7c8bdef8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 13:03:18 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ownslides/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How to set up a free booking page for trial lessons, a guide for language tutors]]></title>
            <link>https://medium.com/@ownslides/how-to-set-up-a-free-booking-page-for-trial-lessons-a-guide-for-language-tutors-f952da55e0ae?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/f952da55e0ae</guid>
            <category><![CDATA[language-learning]]></category>
            <category><![CDATA[teaching]]></category>
            <category><![CDATA[tutoring]]></category>
            <category><![CDATA[education]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 19:45:32 GMT</pubDate>
            <atom:updated>2026-11-15T19:46:27.053Z</atom:updated>
            <content:encoded><![CDATA[<p>If you teach a language online, most students decide in the trial lesson whether to stay. Getting them to that lesson is the hard part. A student finds you on Instagram or gets your name from a friend. Then they have to message you, wait for an answer, work out your time zone and agree on a time. Some give up at each of those steps.</p><p>A booking page removes those steps. The student opens one link, sees who you are, picks a free time in their own time zone and books. You get a notification instead of a week of messages.</p><p>This guide shows how to set one up for free on OwnSlides and what to do with the link once it exists.</p><h3>What a good trial lesson page needs</h3><p>A student deciding on a trial wants four answers, in this order:</p><p>Scheduling tools answer the third question and skip the rest. A link-in-bio page answers the first two and has no calendar. The teacher page on OwnSlides answers all four on one page.</p><h3>Step 1. Make the page</h3><p>Sign up for a free account, open <strong>Create</strong> and pick <strong>Teacher page</strong>. The editor shows a checklist under <strong>Ready to publish</strong>:</p><p>Write the <strong>About you</strong> text for one student, not for everyone. “Conversation-first lessons for adults who understand more than they can say” tells a reader in one line whether you are right for them. “I am a passionate teacher with ten years of experience” does not.</p><p>You can also pick a look for the page, either a chalkboard or paper in one of several colours. Then add a short line that shows on the board, like “Let’s talk from week one.”</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*dHCzpBCMpfJCpBlZ.jpg" /></figure><h3>Step 2. Set the lessons students can book</h3><p>Under <strong>What students book</strong>, add a lesson type. A common setup for a trial is:</p><ul><li>Name: Trial lesson</li><li>Length: 25 or 30 minutes</li><li>Who can book it: Anyone</li><li>Once per student: on, so each new student books one trial</li></ul><p>You can add more lesson types later, for example a lesson that only your current students can book.</p><p>Students pay nothing for lessons booked on OwnSlides. If you also sell paid lessons on your own site, italki or Preply, you can point the page’s main button there and keep the free trial as a second option. Step 5 covers this.</p><h3>Step 3. Add your weekly hours</h3><p>Under <strong>Weekly hours</strong>, set when students can book you. A day can have two ranges, such as a morning block and an evening block. You can copy one day’s hours to the rest of the week and set your <strong>Time zone</strong>.</p><p>Students always see these times in their own time zone, with your local time next to it. International students no longer have to ask what 9:00 means for them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*WmQdX6jzA1uRrjGO.jpg" /></figure><p>Set two rules on day one:</p><p>Under <strong>Lesson video</strong>, paste the link you already use, such as a Zoom, Teams or Google Meet room. Only a student with a booking sees it, and only until that lesson ends.</p><p>If you use Google Calendar, connect it. Two things change:</p><ul><li>OwnSlides stops offering the times you are busy in Google Calendar. It sees only that you are busy, never what you are doing.</li><li>You can have a new Google Meet link made for each lesson, which only that student sees.</li></ul><h3>Step 5. Choose the main button</h3><p>The big button at the top of the page is the first thing a visitor taps. You have three choices:</p><p>If most of your income comes from a marketplace, put that link on the main button and keep the free trial underneath it. If you are building your own list of students, make the free trial the main button.</p><p>Below the button you can add your other links, like Instagram, TikTok, YouTube and your website. The page then works as your link in bio.</p><p>Press <strong>Publish page</strong>. Your page now has its own address, and the <strong>Share</strong> sheet has five ways to spread it:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/898/0*Gop1sUBuWzx-hRJK.jpg" /></figure><p>You can also list yourself in the <strong>Teacher directory</strong>, where people looking for a teacher of your language can find you.</p><h3>What the student sees</h3><p>A student who opens your link sees your photo, your language, your About line and the main button.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*qVG-wQbq8j3oc2LR.jpg" /></figure><p>They tap the trial lesson, pick a day and a time in their own time zone, and can add a note about their level or what they want to practise.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*4pPzlachvIgxpSsz.jpg" /></figure><p>To book, they make a free account. They then get a confirmation, or a “request sent” message if you approve each booking yourself. They also get an <strong>Add to calendar</strong> button for Google, Apple or Outlook and a reminder an hour before the lesson. During the lesson, the video link is on the lesson page and in their notifications.</p><p>You get the booking right away, with their note, and the lesson appears on your schedule.</p><h3>After the trial, give them a reason to come back</h3><p>On the same page, under <strong>Learn with me</strong>, students see your public decks, notes, shorts and exercise sets. After the trial you can give a new student homework from those and see whether they practised before the second lesson. A student who has already practised with your material between lessons has a concrete reason to book again.</p><h3>Quick checklist</h3><ul><li>A real photo and a one-line About written for the student you want</li><li>One trial lesson type, 25 to 30 minutes, once per student</li><li>Weekly hours and your time zone, with a few hours’ notice</li><li>A video link, and Google Calendar connected if you use it</li><li>The main button pointing where most of your students should go</li><li>The link in your Instagram, TikTok and YouTube bios and in your email signature</li></ul><p><a href="https://ownslides.com/en/features/teacher-page">Make your teacher page</a>. It is free, and you can take it down at any time.</p><p><em>Originally published at </em><a href="https://ownslides.com/en/blog/free-booking-page-for-language-tutors"><em>https://ownslides.com</em></a><em> on November 15, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f952da55e0ae" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to make short language-learning videos for Reels and TikTok without editing skills]]></title>
            <link>https://medium.com/@ownslides/how-to-make-short-language-learning-videos-for-reels-and-tiktok-without-editing-skills-567f3da260b6?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/567f3da260b6</guid>
            <category><![CDATA[tik-tok]]></category>
            <category><![CDATA[teaching]]></category>
            <category><![CDATA[language-learning]]></category>
            <category><![CDATA[content-creation]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 19:45:26 GMT</pubDate>
            <atom:updated>2026-11-15T19:48:17.240Z</atom:updated>
            <content:encoded><![CDATA[<p>If you teach a language, short videos are the cheapest way for new students to find you. One clear 40-second video about a mistake every learner makes can reach more people than a month of posts. The problem is time. Writing, recording, cutting, adding captions and fixing the timing take hours, and many teachers stop after a few videos.</p><p>This guide shows a faster way. You build the video as a set of slides, the way you would build a lesson, and export it as a vertical video. You never touch a timeline.</p><h3>What makes a language video work</h3><p>The language videos that people save and share usually have the same shape:</p><p>These formats work well for language teachers:</p><h3>Step 1. Start a short</h3><p>On OwnSlides, a short is a slide lesson you can post as a video or give to your students. Open <strong>Create</strong>, pick <strong>Shorts</strong> and press <strong>Make a short</strong>. Choose the language you teach and your audience’s language, give the short a name, and the editor opens.</p><p>You can write every slide yourself or have the editor draft them. Press <strong>Add Slide</strong> and choose <strong>Generate Short</strong>. Describe the topic the way you would explain it to a colleague, for example <em>“the difference between ser and estar for A2 learners, 45 seconds”</em>. You can put the number of slides, the length and the level in the description.</p><p>Then pick a <strong>Short style</strong>. The styles match the formats above: Grammar Myths, Slang &amp; Informal, Cultural Differences, Real Conversations, False Friends, Say It Like a Native, Song Lyrics, and Classic Short, which mixes vocabulary, grammar and practice. For Song Lyrics, paste the lyrics and the short uses the real words.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NMNaIvOoCYDpTkb85pvVrA.jpeg" /></figure><p>The form has a few more options:</p><h3>Step 2. Check every slide</h3><p>The draft can contain mistakes, so read every slide as a teacher would. Change any example that sounds unnatural, shorten any line that would not fit on a phone, and cut slides that repeat.</p><p>Two rules help slides read well on a phone:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*JZuGlbrZb-mOlPrs.jpg" /></figure><h3>Step 3. Add voice and music</h3><p>For each slide you can add:</p><p>If you teach pronunciation, record yourself. Viewers hear how you speak and explain before they ever book a lesson.</p><h3>Step 4. Export the video</h3><p>Open the menu with three dots, choose <strong>Export video or pictures</strong> and keep the output as a single MP4 video file. If you would rather post a carousel, you can export the slides as separate clips or images.</p><p>On the free plan, a video can have up to 10 slides and shows a small watermark with your social handle, so viewers can find your account. With Plus you can export up to 30 slides in Full HD, without the OwnSlides mark.</p><p>The export runs on our servers, so you can close the tab and come back when it is ready.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/898/0*cBZZIsXUOiM-CU2T.jpg" /></figure><h3>Step 5. Post it, then use it again</h3><p>Post the video to Instagram Reels, TikTok and YouTube Shorts. The same vertical video works on all three. In the caption, repeat the idea in one sentence and ask a question that invites a comment, like “Which one did you learn at school?”</p><p>The same short is also a lesson your students can watch and practise with. You can give it to one student or a whole class, with a deadline if you like, and see who watched it. If you have a <a href="https://ownslides.com/en/blog/free-booking-page-for-language-tutors">teacher page</a>, your public shorts show there too. A viewer who likes your video can watch more and book a trial lesson.</p><h3>A simple posting plan</h3><p>Picking the topic in advance saves a decision every day. One week could look like this:</p><ul><li>Monday, a false friend</li><li>Wednesday, one phrase written formally and then casually</li><li>Friday, a mistake your students made this week, without names</li></ul><p>At about fifteen minutes per video, three videos a week is a pace most teachers can keep.</p><h3>Quick checklist</h3><ul><li>One idea per video and a hook on the first slide</li><li>30 to 60 seconds, and no more than 10 slides on the free plan</li><li>Every slide read and corrected by you</li><li>Your voice on at least one slide</li><li>The same video on Reels, TikTok and Shorts</li><li>The short given to your students too</li></ul><p><a href="https://ownslides.com/en/features/create">Make your first short</a>. It is free to make and publish.</p><p><em>Originally published at </em><a href="https://ownslides.com/en/blog/how-to-make-language-learning-videos-for-reels-and-tiktok"><em>https://ownslides.com</em></a><em> on November 15, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=567f3da260b6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to make a vocabulary deck for your students in five minutes]]></title>
            <link>https://medium.com/@ownslides/how-to-make-a-vocabulary-deck-for-your-students-in-five-minutes-818e78975491?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/818e78975491</guid>
            <category><![CDATA[vocabulary]]></category>
            <category><![CDATA[teaching]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[language-learning]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 19:45:09 GMT</pubDate>
            <atom:updated>2026-11-15T19:49:11.914Z</atom:updated>
            <content:encoded><![CDATA[<p>You teach twenty new words on Tuesday, and on Thursday half the class remembers four of them. Between the two lessons, most students never looked at the words again.</p><p>A vocabulary deck closes that gap. Students practise a few minutes a day, and each word comes back just before they would forget it. This guide shows how to make a deck for your students in about five minutes, give it as homework and check who practised.</p><h3>Why a deck works better than a word list</h3><p>Students read a list once. They practise a deck in short sessions over several days, and the words they get wrong come back sooner than the ones they know. This method is called spaced repetition, and researchers have studied it for decades. We summarised <a href="https://ownslides.com/en/blog/spaced-repetition-research">what the research says</a>.</p><p>A deck also helps you as a teacher:</p><h3>Step 1. Decide what goes in</h3><p>A good deck for one week has 10 to 25 words from one topic or one lesson. Pick words your students will meet again soon, in your next lesson, in a text you will read together or in a conversation you will practise.</p><p>You can start from:</p><p>Your own list usually works best, because students learn the words you will actually use in class.</p><h3>Step 2. Make the deck</h3><p>On OwnSlides, open <strong>Create</strong>, pick <strong>Decks</strong> and start a new deck. In the deck editor, press <strong>Add</strong>. You get four ways to fill it:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*IXEOReZgD6p8bwbA.jpg" /></figure><p>For a new lesson, <strong>Several from a list or topic</strong> is the fastest. Pick the language for the meanings, then choose <strong>Topic</strong>, <strong>My list</strong> or <strong>Surprise me</strong>, and pick a level or let it match the learner’s level. You get up to 25 words, each with a meaning, pronunciation and an example sentence. OwnSlides skips words already in your collection, and a narrow topic gives fewer words instead of filling the deck with words nobody needs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*KOW0uj89SFIt7y-h.jpg" /></figure><p>Your first deck each day is built for free on every plan, and you can always type the cards yourself.</p><h3>Step 3. Check the cards</h3><p>Read the deck the way you would read a student’s homework, and look for these:</p><p>A new deck is a private draft. Nobody sees it until you publish it or give it to students.</p><h3>Step 4. Give it to your students</h3><p>When the deck is ready, give it to one student or a whole class. You can add a deadline, for example the day before your next lesson.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/898/0*_qfLmImFzG5ECu2J.jpg" /></figure><p>Students practise on their phone or computer for a few minutes a day. Each word comes back on its own schedule. A student who already knows “la cuenta” won’t see it ten times, and a student who keeps forgetting “el camarero” will see it more often.</p><p>You can also attach a deck to a note or a short, so students practise the words next to the explanation.</p><h3>Step 5. Check who practised</h3><p>Before the next lesson, open the deck and see who opened it, who finished it and how many students practised this week. Use that to plan how the lesson starts. If most of the class practised, open with a conversation that uses the words. If few did, start with a quick game and move the conversation later.</p><p>After a few weeks you will also see which students practise every week and which practise only before a test. Asking a student about that pattern gets a more honest answer than “did you study?”</p><h3>Already using Anki or Quizlet?</h3><p>Your students don’t have to start over. They can import Anki decks in .apkg or .colpkg format, CSV files and Quizlet exports. OwnSlides keeps their review history, so a card they have known for 45 days stays known.</p><h3>A five-minute routine after every lesson</h3><ol><li>Right after the lesson, paste the new words into a deck. This takes about 2 minutes.</li><li>Check the meanings and examples, about 2 minutes.</li><li>Give the deck to the class with a deadline before the next lesson, about 1 minute.</li></ol><p>With these five minutes after each lesson, your students arrive at the next one already knowing the words.</p><h3>Quick checklist</h3><ul><li>10 to 25 words from one topic or one lesson</li><li>The meaning you taught, natural examples and the right level</li><li>Given to the class with a deadline before the next lesson</li><li>Checked before the lesson to see who practised</li><li>The lesson plan adjusted to what you see</li></ul><p><a href="https://ownslides.com/en/features/create">Make your first deck</a>. It is free to make and to give to your students.</p><p><em>Originally published at </em><a href="https://ownslides.com/en/blog/how-to-make-a-vocabulary-deck-for-your-students"><em>https://ownslides.com</em></a><em> on November 15, 2026.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=818e78975491" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your Language Teaching Business Doesn’t Need More Content. It Needs a Timer.]]></title>
            <link>https://medium.com/@ownslides/your-language-teaching-business-doesnt-need-more-content-it-needs-a-timer-5e86887c30a3?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/5e86887c30a3</guid>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[solopreneur]]></category>
            <category><![CDATA[online-teaching]]></category>
            <category><![CDATA[language-teaching]]></category>
            <category><![CDATA[content-creation]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Mon, 03 Aug 2026 22:40:01 GMT</pubDate>
            <atom:updated>2026-08-03T22:40:01.642Z</atom:updated>
            <content:encoded><![CDATA[<p>Two hours a week is enough. Here’s the budget — and what to cut to afford it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1YzXKZqCubuUgK2c" /><figcaption>Photo by <a href="https://unsplash.com/@windows?utm_source=medium&amp;utm_medium=referral">Windows</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>There’s a version of teaching languages online that nobody signed up for. Sunday evening in a design tool. Tuesday fighting a caption. Thursday wondering why the reel flopped. And somewhere in between, you also teach.</p><p>The marketing ate the job.</p><p>My answer is two hours a week. Not two hours a day. Two hours a week, spent on a system instead of improvised at 6pm. If you’re spending more than that, the extra hours aren’t going into anything a student ever sees. They’re going into production and indecision.</p><p>Let me show you the budget, and then let me show you what it costs you not to have one.</p><h3>First, the part nobody says out loud</h3><p>You’re not spending ten hours a week on marketing because you’re inefficient.</p><p>You’re spending ten hours because when you’re not visible, you feel like you’re not working. Because a quiet week feels like proof that you’re not really a business. Because the feed is the only place where the effort is legible — a lesson happens once, in private, and then it’s gone, but a post sits there looking like evidence.</p><p>Content production expands to fill available anxiety.</p><p>That’s the actual mechanism. And it’s why productivity advice bounces off: you can’t optimise your way out of a feeling. You need a boundary instead.</p><h3>What the extra hours actually cost</h3><p>Say you spend ten hours a week on marketing and you charge 30 euros a lesson.</p><p>Eight of those hours are unpaid overhead. That’s 240 euros a week you’re not billing, or roughly 960 euros a month, or the better part of a five-figure sum across a year — spent making content whose whole purpose was to sell the lessons you no longer had time to teach.</p><p>Adjust the rate to yours. The arithmetic goes the same direction every time.</p><h3>The two-hour week</h3><p>40 minutes — batch-produce the week’s posts. Three posts a week is plenty. Making all three in one sitting is dramatically faster than three separate sittings, because the setup cost — tool open, head in content mode — gets paid once instead of three times. Whatever tool you use, the machine part can run while you do something else. What takes the time is deciding what to say, which is exactly why you want to decide it once.</p><p>20 minutes — captions and scheduling. Write all three captions together, schedule them, close the tab. Scheduled posts kill the “it’s 6pm, I should post something” panic, which reliably produces the worst content you will ever make.</p><p>30 minutes — engagement, split across the week. Ten minutes, three times: answer comments, reply to messages, comment on a few accounts in your niche. This half hour usually beats an extra post. Students hire teachers who talk to them.</p><p>30 minutes — one experiment. One new thing per week. A different hook, a quiz slide, a new series name. Glance at last week’s numbers first, but don’t drown in them — at three posts a week, your sample sizes are noise for months. The experiment habit is what compounds, not the analytics.</p><p>That’s 120 minutes. Past that, returns fall off a cliff until you’re big enough to pay someone else to do it.</p><h3>What to cut to afford it</h3><p>Cut original design. All your posts should share one template family and one theme. Choosing fonts weekly isn’t marketing. It’s procrastination with a creative alibi.</p><p>Cut platforms. One platform done consistently beats three done spasmodically. Pick where your students already are. Repurpose elsewhere only when it’s free — an export button, not a redesign.</p><p>Cut trends that don’t teach. Dancing to a trending sound might reach people. It won’t reach people who came for the subjunctive. Reach without fit is a vanity spike.</p><p>Cut pixel-level perfectionism. A student choosing between two teachers is comparing clarity, warmth, and evidence that you show up reliably. Nobody has ever booked a lesson because of your kerning.</p><h3>You’re allowed to do less</h3><p>Here’s the permission slip, since nobody hands these out.</p><p>You are allowed to post three times a week. You are allowed to reuse the same template for a year. You are allowed to have a quiet week and not interpret it as a verdict on your career.</p><p>Marketing exists to support your teaching. The moment it costs more energy than the lessons it’s supposed to sell, it has stopped doing its job — and no amount of extra effort fixes that, because effort is the problem.</p><p>Teach first. Market on a timer.</p><p><em>I built </em><a href="https://ownslides.com/"><em>OwnSlides</em></a><em> because the 40-minute block was the one I kept blowing past. It’s template-first: pick a theme once, and a ten-slide deck gets generated in about 20 seconds — the deciding-what-to-say part, which is the part that actually eats your Sunday. Rendering the finished carousel takes a few minutes more, but it runs on its own, so you can close the tab and go teach. If your two hours keep becoming six, that’s the block to fix first.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5e86887c30a3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An AI Language Tutor Is a Gym, Not a Match: 5 Tests to Tell if Yours Is Any Good]]></title>
            <link>https://medium.com/@ownslides/an-ai-language-tutor-is-a-gym-not-a-match-5-tests-to-tell-if-yours-is-any-good-8c8560aaa379?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/8c8560aaa379</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[edtech]]></category>
            <category><![CDATA[learning]]></category>
            <category><![CDATA[language-learning]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Sun, 26 Jul 2026 09:39:51 GMT</pubDate>
            <atom:updated>2026-09-04T15:11:36.743Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*2lKb6Ju7o4S6vDcI" /><figcaption>Photo by <a href="https://unsplash.com/@towfiqu999999?utm_source=medium&amp;utm_medium=referral">Towfiqu barbhuiya</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>An AI tutor will never give you the social stakes of a real conversation. There’s no fear of being laughed at, no reward for landing a joke in a second language, no stranger deciding whether you’re worth the effort of slowing down for. That’s not a flaw to be engineered away. It’s the category.</p><p>An AI tutor is a gym. The point of the gym is to make the match survivable.</p><p>Which means the question isn’t “can it replace a human tutor?” It can’t. The question is whether it’s a good gym or a room with mirrors. Most people picture an AI language tutor as a chatbot that compliments your typos, and that image is a few years out of date. It also makes learners skip the most patient teacher they have access to.</p><p>I’ve built one of these: OwnSlides, which has a <a href="https://ownslides.com/en/features/chat">text</a> and <a href="https://ownslides.com/en/features/voice">voice</a> tutor. So rather than argue in the abstract, here are five tests. Each takes about two minutes. Run them on whatever tool you’re evaluating, mine included. I’ll point out where OwnSlides does something specific, and where the problem is just genuinely hard for everyone.</p><h3>Test 1: Say something slightly beyond your level. Does the difficulty move?</h3><p><strong>What to do:</strong> Answer two or three questions in the most confident, complex sentences you can manage. Then, deliberately, stall, giving a one-word answer, or say you don’t understand.</p><p><strong>What good looks like:</strong> The difficulty climbs when you’re fluent and drops when you stall, and it never announces itself. You shouldn’t see the machinery. You should only notice that the conversation stays slightly harder than comfortable.</p><p><strong>What it means if it fails:</strong> A tutor that talks to a beginner in native-speed idioms is showing off. One that talks to a B2 learner like a toddler has stopped paying attention. Both are the same failure, a fixed difficulty that ignores you.</p><p>That band of “slightly too hard” has a name in learning science. Lev Vygotsky called it the <a href="https://www.simplypsychology.org/zone-of-proximal-development.html">zone of proximal development</a>: the range between what you can do alone and what you can do with help. It’s where the learning actually happens. Too easy is entertainment. Too hard is noise.</p><p>Building this into OwnSlides was harder than I expected, and the reason is instructive: the tutor has to infer your level from very little evidence, early, and be willing to revise that guess mid-conversation. Get it wrong in the confident direction and you drown a beginner. Get it wrong in the cautious direction and you bore someone who was ready for more.</p><h3>Test 2: Make a real mistake. Does practice follow the mistake, or the schedule?</h3><p><strong>What to do:</strong> Mix up two tenses mid-conversation. Use the wrong preposition twice. Then keep talking and see what the tool does with it.</p><p><strong>What good looks like:</strong> The conversation spawns a short, targeted drill on exactly that confusion, then returns to the conversation. Brief, earned, specific.</p><p><strong>What it means if it fails:</strong> There are two opposite failure modes here, and most tools pick one. Language apps default to the endless quiz: tap the matching card, fill the blank, collect the confetti, none of it connected to anything you actually got wrong. Pure chatbots default to the opposite. Pleasant, frictionless talk with no pressure to produce anything precise. You leave feeling good and nothing sticks.</p><p>The format of the exercise matters much less than its trigger. Practice should respond to your errors, not arrive on a timer.</p><p>In OwnSlides this is the piece I’ve rewritten most often. The tutor generates the drill from the mistake you just made, which sounds obvious and is a genuine engineering problem. You’re deciding, live, whether an error is worth interrupting for, then building a two-minute exercise around it, then getting back to the conversation before the thread goes cold.</p><h3>Test 3: Say something understandable but imperfect. Does it interrupt you?</h3><p><strong>What to do:</strong> Drop an article. Get a gender wrong. Say something that’s clearly comprehensible and clearly not correct, then watch whether it addresses your point or your grammar first.</p><p><strong>What good looks like:</strong> It answers what you said. Then it circles back with one small correction. Errors that block communication get fixed immediately; errors of polish get noted and drilled later. That’s triage, and it’s what a good human tutor does automatically.</p><p><strong>What it means if it fails:</strong> If it stops the conversation over a missing article, it’s optimising for looking smart rather than for you getting better.</p><p>This isn’t a style preference. <a href="https://www.cambridge.org/core/journals/studies-in-second-language-acquisition/article/abs/corrective-feedback-and-learner-uptake/59229F0CA2F085F5F5016FB4674877BF">Lyster and Ranta’s classic study</a> of French immersion classrooms found that the most common teacher move, the recast that quietly repeats your sentence back in corrected form, was also the <em>least</em> likely to produce any actual repair from the student. Correction that’s frictionless to deliver is often the kind that doesn’t land. Separately, work on foreign language anxiety has long linked harsh or relentless correction to learners withdrawing from speaking at all. A silent learner learns nothing.</p><h3>Test 4: Come back tomorrow. Does it remember today?</h3><p><strong>What to do:</strong> Have a session. Note two or three words you struggled with. Come back the next day and see whether they surface.</p><p><strong>What good looks like:</strong> Thursday’s conversation is shaped by Tuesday’s difficulties. The words you fumbled come back, in a review deck, in a drill, or just seeded into the next conversation.</p><p><strong>What it means if it fails:</strong> A session that starts from zero every time is a demo, not a tutor. Memory is also what makes progress visible, and visible progress is what keeps the habit alive when motivation dips. A skills chart moving, a review deck growing, an accuracy trend going the right way. None of that is gamification garnish. They’re the reason you open the app on day forty.</p><p>OwnSlides handles this by pushing words you’ve struggled with into a spaced-repetition deck, so they come back at widening intervals rather than at random. That’s a well-understood technique and not remotely unique to me, but it does require the tutor to be writing things down about you during the conversation, which is the part a stateless chatbot structurally cannot do.</p><h3>Test 5: Ask it what you’re bad at. Does it have an answer?</h3><p><strong>What to do:</strong> Straightforwardly ask: what am I struggling with?</p><p><strong>What good looks like:</strong> A specific answer. “Past-tense endings” or “you avoid the subjunctive”, something derived from what you’ve actually done, that you could act on today.</p><p><strong>What it means if it fails:</strong> Vague encouragement means it isn’t tracking anything. “You’re doing great, keep practising!” is what a tool says when there’s no model of you underneath it. If it can’t tell you what you’re bad at, it can’t be building anything around your weaknesses either. Tests 1, 2 and 4 were all quietly depending on that model existing.</p><h3>The shape of a good session</h3><p><em>(Illustrative. This is the pattern the tests above are describing, not a transcript of any particular product.)</em></p><blockquote><strong><em>You:</em></strong><em> Yesterday I go to the market and buy some fish.</em></blockquote><blockquote><strong><em>Tutor:</em></strong><em> Nice, what kind of fish? Anything you hadn’t cooked before?</em></blockquote><blockquote><strong><em>You:</em></strong><em> Is a fish called dorada. I never cook it before.</em></blockquote><blockquote><strong><em>Tutor:</em></strong><em> Dorada’s a good one, a sea bream. Quick thing before you tell me how it turned out: you said “I go” and “I never cook it.” Both should be past. Try that first sentence again?</em></blockquote><blockquote><strong><em>You:</em></strong><em> Yesterday I went to the market and bought some fish.</em></blockquote><blockquote><strong><em>Tutor:</em></strong><em> That’s it. So how did the dorada turn out?</em></blockquote><p>Four things happened there. The tutor answered the content before the grammar. It let the first error pass to keep the conversation moving, then caught the pattern when it repeated. The correction was one line, not a lecture. And the conversation resumed exactly where it left off.</p><p>That’s the whole thing. Everything else is interface.</p><h3>Frequently asked</h3><p><strong>Can an AI tutor replace a human one?</strong> No, and the useful version of that answer is specific: it can’t replicate social stakes, cultural nuance, or the accountability of a person expecting you on Thursday. What it can do is give you unlimited, unembarrassing repetition, which is exactly what most learners are short of between human sessions.</p><p><strong>Is talking to an AI actually useful if it never gets impatient?</strong> That’s the trade. Infinite patience is why you’ll say things to it you’d be too self-conscious to attempt with a person. It also means the pressure has to come from the tool’s design, meaning difficulty that climbs and corrections that land, rather than from social risk. That’s what Tests 1 and 3 are checking for.</p><p><strong>How much does this beat just using a general-purpose chatbot?</strong> For pure conversation practice, a general chatbot does fine. It fails Tests 4 and 5, with no memory across sessions and no model of your weaknesses, so it can’t build anything cumulative. If you only ever want to talk, that may not matter to you.</p><h3>What this all costs you if you get it wrong</h3><p>Nothing dramatic, and that’s the problem. A bad AI tutor doesn’t fail loudly. It gives you a pleasant twenty minutes a day, a streak counter, and a slowly growing certainty that you’re not really improving. The tests above take ten minutes total and will tell you inside the first session which one you’ve got.</p><p>I build <a href="https://ownslides.com/">OwnSlides</a>, text and voice tutoring built along exactly these lines, because these are the things I kept wanting and not finding. Run the five tests on it. Run them on whatever else you’re considering. If a product fails them, the logo doesn’t matter.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8c8560aaa379" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The 15-Minute Daily Language Routine That Survives Real Life]]></title>
            <link>https://medium.com/@ownslides/the-15-minute-daily-language-routine-that-survives-real-life-6a80c24cefad?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/6a80c24cefad</guid>
            <category><![CDATA[discipline]]></category>
            <category><![CDATA[apps]]></category>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[language-learning]]></category>
            <category><![CDATA[language]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Thu, 23 Jul 2026 15:34:46 GMT</pubDate>
            <atom:updated>2026-09-04T15:48:18.692Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*vTY_qBRhIfSkmfqb" /><figcaption>Photo by <a href="https://unsplash.com/@ilyuza?utm_source=medium&amp;utm_medium=referral">Ilyuza Mingazova</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Every January the same thing happens on every language platform: sign-ups spike, streaks bloom, and by week three the graveyard fills. The learners who survive are usually not the motivated ones. They are the ones whose routine costs less than their worst day has energy for.</p><p>That is the design constraint most study plans ignore. A 45-minute evening block works until the first evening you have a headache, and then the streak breaks, and broken streaks tend not to restart. So the routine below is built backwards from the worst day rather than the best one: fifteen minutes, three blocks, each with a reason to exist.</p><h3>The routine</h3><p>Minutes 1 to 3: the anchor. Play a <a href="https://ownslides.com/games">daily word game</a>. Habits attach to cues, and the strongest cues are habits you already have: the morning coffee, the commute, the scroll you were doing anyway. A daily puzzle is the easiest possible entry: two minutes, mildly competitive, resets at midnight whether you show up or not. You are not studying, you are playing, and that distinction is load-bearing. Starting is the whole battle; nobody quits at minute four.</p><p>Minutes 3 to 8: the compounding block. <a href="https://ownslides.com/en/features/my-words">Review your saved words</a>. Spaced repetition is the highest-return five minutes in language learning. The input queue comes free from the anchor, the words the game handed you yesterday and the ones you got wrong. Words with a story attached stick better than words from a list.</p><p>Minutes 8 to 15: the production block. <a href="https://ownslides.com/en/features/chat">Say or write something</a>. Recognition without production is how people end up reading menus they cannot order from. Seven minutes of output: a short conversation, a voice note narrating your day, a paragraph written and self-corrected. Rotate freely. The only rule is that something leaves your head in the target language every day.</p><p>No block depends on willpower arriving. Each is cued by finishing the previous one, and the first one is a game.</p><h3>Why this survives when bigger plans die</h3><p>It passes the worst-day test. On a terrible day you do the anchor only, three minutes, and the habit survives. The consistency of showing up matters more than the size of the session, because the habit is the asset. Fifteen minutes daily is roughly 90 hours a year, not a full CEFR level at the higher stages, but a serious amount of contact time from something that fits inside a coffee.</p><p>The streak is a tool, not a master. Streaks motivate until the first accident, then they demoralise. Two rules keep them healthy: a daily minimum small enough to defend at 11:50pm, and forgiveness for a genuinely bad day. A routine that punishes the flu is a badly designed routine.</p><p>Same time, same trigger. Habits bind to context. The routine after the morning coffee survives; the routine “sometime today” dies of freedom. Pick the trigger once and stop renegotiating it.</p><p>Visible <a href="https://ownslides.com/en/features/progress">progress</a>, weekly. Daily progress is imperceptible, and that alone kills motivation. Once a week look at something cumulative: words in your deck, a skill chart, last month’s recording against today’s.</p><h3>Building it tonight</h3><p>Pick your trigger: coffee, commute, lunch queue. Pick one daily game as the anchor. Pick where saved words go. Pick your default production act. Write the four choices down, and on bad days defend only the first three minutes.</p><p>Everything here exists somewhere free, and the architecture works with index cards and a podcast.</p><p><em>I build </em><a href="https://ownslides.com/"><em>OwnSlides</em></a><em>, where the daily games feed the review deck and a tutor covers the production block.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6a80c24cefad" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Tell If Your Language Is Actually Improving]]></title>
            <link>https://medium.com/@ownslides/how-to-tell-if-your-language-is-actually-improving-d891dbfb2568?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/d891dbfb2568</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[language-learning]]></category>
            <category><![CDATA[self-improvement]]></category>
            <category><![CDATA[language]]></category>
            <category><![CDATA[language-app]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Wed, 22 Jul 2026 18:16:01 GMT</pubDate>
            <atom:updated>2026-09-04T15:48:55.334Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*95gK65wMpJT2ZeLK" /><figcaption>Photo by <a href="https://unsplash.com/@kimberlyfarmer?utm_source=medium&amp;utm_medium=referral">Kimberly Farmer</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>The cruel thing about language learning is that daily progress is imperceptible. You study for three weeks, feel identical, and conclude it is not working. Meanwhile the only number most apps show you is a streak, which measures showing up rather than improving. A 200-day streak of easy reviews is 200 days of maintenance.</p><p>Here are five signals that track improvement instead, each with a way to check it yourself.</p><h3>Production, not recognition</h3><p>Multiple-choice scores flatter you, because the answer is already on the screen. The honest metric is what you can generate from nothing. Track the share of review items you can produce, in writing or out loud, rather than merely recognise. If your tool separates the two modes, watch the production number. If it does not separate them, that itself is worth knowing.</p><p>Doing it by hand: once a week, take ten words you “know” and write a sentence with each, no peeking. Count the clean ones. That number moving is real progress.</p><h3>The same task, re-measured</h3><p>Improvement only shows against a fixed baseline. Pick a repeatable task, such as describing your kitchen out loud for sixty seconds or reading the same short paragraph, and record it monthly. Month-old you is the only fair opponent you will ever have.</p><p>This is why any scored task doubles as a progress log. The score on a given day matters less than the trend.</p><h3>Skill spread, not skill total</h3><p>“How good is your Spanish” is four questions wearing one coat: listening, reading, speaking, writing. Everyone develops lopsided, and the lopsidedness stays invisible until you chart it. Four axes, honest 1 to 10, drawn by hand is enough. It shows in one glance what a single score hides: the reading spike, the speaking crater.</p><p>The action is always the crater. Time spent on your weakest skill moves the whole system, because the skills feed each other.</p><h3>Time to retrieval</h3><p>Fluency, mechanically, is retrieval speed. Early on you translate in your head; later the words arrive on their own. You can feel that transition, and you can also watch it: in review sessions, notice the hesitation before an answer. Shrinking hesitation on old words is about as direct a fluency measure as exists.</p><p>The milestone worth marking is the first time you catch yourself understanding a sentence without translating it. Write the date down.</p><h3>The size of your active world</h3><p>Count the situations you can handle, not the words you know. Last month: ordering coffee. This month: ordering coffee and complaining that it is cold. The CEFR levels are exactly this, a ladder of situations, and I <a href="https://medium.com/p/42c3b6a52492">wrote a plain-language guide to them</a> separately. Self-assess quarterly against situations, because it is the only measure that answers the question you actually care about, which is whether you can use the language yet.</p><h3>The weekly ritual</h3><p>Daily, look at nothing. Day-to-day noise will only demoralise you. Weekly, spend five minutes on production accuracy and one re-measured task. Monthly, the recording. Quarterly, the situations list.</p><p>A notebook does all of this. The point is the ritual: measure weekly, and invisible progress becomes visible exactly often enough to keep you going.</p><p><em>I build </em><a href="https://ownslides.com/"><em>OwnSlides</em></a><em>, where review modes, skill charts and pronunciation scores are tracked automatically.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d891dbfb2568" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Build a Spaced Repetition System. Most of What You Read About It Is Wrong.]]></title>
            <link>https://medium.com/@ownslides/i-build-a-spaced-repetition-system-most-of-what-you-read-about-it-is-wrong-0317bd594579?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/0317bd594579</guid>
            <category><![CDATA[psychology]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[language-learning]]></category>
            <category><![CDATA[learning]]></category>
            <category><![CDATA[self-improvement]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Wed, 22 Jul 2026 14:14:24 GMT</pubDate>
            <atom:updated>2026-09-04T15:47:29.849Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="A page of handwriting half-obscured by the striped shadow of window blinds, some words legible and some lost." src="https://cdn-images-1.medium.com/max/1024/0*hjNJ4T8tVoLT_dWO" /><figcaption>Forgetting is partial and it flattens out. The popular version of the curve gets both wrong. Photo by <a href="https://unsplash.com/@micahboswell?utm_source=medium&amp;utm_medium=referral">Micah Boswell</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Spaced repetition is the closest thing language learning has to a free lunch. Review a word just before you would have forgotten it, and it sticks for longer; repeat, and the intervals stretch out until the word is simply yours. Every vocabulary app is built on this, mine included.</p><p>The underlying effect is real and very well evidenced. Almost everything layered on top of it, meaning the specific numbers, the curve and the clever scheduling, is shakier than the confident tone of most explainers suggests. Some of it is wrong. One piece of it undermines the design of my own product, which is the part I found hardest to write.</p><h3>The forgetting curve is not what you were told</h3><p>You have seen the claim: you forget 50% within an hour, 70% within a day, 90% within a week. It appears in a thousand blog posts and a fair number of pitch decks.</p><p>Hermann Ebbinghaus ran the experiments behind it in 1879 and 1880, publishing in 1885. Three things about that work rarely survive the retelling. He was his own and only subject. He memorised nonsense syllables, which are the fastest-forgetting material anyone has ever constructed. He found meaningful material, like poetry, roughly ten times easier. And his y-axis was not “percentage remembered” at all; it was savings, the reduction in time needed to relearn the list.</p><p>His actual figures: about 58% savings after twenty minutes, 44% after an hour, 34% after a day, 25% after six days, 21% after thirty-one days. The “90% in a week” number is not a misreading of Ebbinghaus. It does not appear in his data at all, and I could not trace it to any primary source.</p><p>Notice also what the real numbers do. Between six days and a month, the loss is about four percentage points. The curve flattens hard. It is not a slide toward zero, which is what the popular version implies.</p><p>A 2015 replication by Murre and Dros reproduced the general shape, and was also a single-subject study, a detail nearly every write-up omits. Their own caution is worth more than their confirmation: the famous curve may be partly an averaging artefact rather than one clean law of decay.</p><h3>What is actually solid</h3><p>Two findings survive scrutiny comfortably, and they are the two that matter.</p><p>Spacing beats cramming. Cepeda and colleagues reviewed 184 articles and 317 experiments in 2006 and found spaced practice produced about 47% recall against 37% for massed practice. Ebbinghaus himself had noticed it: 38 repetitions spread over three days did as much good as 68 repetitions crammed into the day before.</p><p>Testing beats re-reading. This one is larger than most people expect. Karpicke and Roediger tested Swahili-English word pairs in 2008; a week later, learners who had practised retrieving the words remembered about 80%, against roughly 35% for those who had restudied them. The learning curves during the session were identical. The difference only appeared later, which is exactly why it is so easy to miss.</p><p>The metacognitive twist is the useful part for a learner. In those studies, participants consistently predicted they would do better with the method that worked worse. Re-reading feels productive; retrieval feels like failure, because you are confronted with what you cannot recall. The uncomfortable one is the one that works.</p><p>When a review of learning techniques by Dunlosky and colleagues in 2013 scored ten popular study methods, only two came out as high utility: distributed practice and practice testing. A flashcard review session is both at once. That is the entire case for spaced repetition, and it is a strong one.</p><h3>The part that makes my algorithm look silly</h3><p>Every SRS app, mine included, is built on expanding intervals. Get a word right and the gap grows; get it wrong and it collapses. My implementation is SM-2, the algorithm behind Anki and most of its descendants. In my code a word you keep getting right is scheduled at one day, then three, then the interval multiplied by an ease factor that starts at 2.5 and can never fall below 1.3. In practice that is roughly 1, 3, 8, 20, 50, 125 days.</p><p>That expansion is the part everyone finds satisfying. It is also the part with almost no evidence behind it.</p><p>A 2021 meta-analysis by Latimier and colleagues compared spacing against cramming and found a large effect, as expected. Then it compared expanding intervals against simply spacing reviews evenly, and found essentially nothing. Karpicke and Roediger had found the same thing in 2007, where equal spacing actually beat expanding schedules at a two-day delay.</p><p>SM-2 itself deserves a look. The founding paper by Wozniak and Gorzelańczyk in 1994 had seven subjects and no control group. It was parameter-fitting, not validation. As far as I can find, no independent study has ever validated Anki’s scheduler, or FSRS, or mine, against a simpler alternative.</p><p>So the honest summary is this: spacing your reviews is extremely well supported. Retrieving rather than re-reading is extremely well supported. The specific escalating schedule your app is so proud of, mine included, is a plausible engineering guess dressed in the credibility of the two findings underneath it.</p><h3>What the research says to do instead</h3><p>The most practically useful study here is Cepeda and colleagues in 2008, which put 1,354 people through trivia learning at varying gaps. The optimal gap depends on how long you want to remember something, and the ratio shrinks as the horizon grows: to remember something for a week, review after about a day; for a month, after around eleven days; for a year, after about three weeks.</p><p>Two things follow. The popular “always space at 20% of your target” rule is an extrapolation the authors themselves declined to make. And there is a useful asymmetry, in their words: the cost of a gap that is too long is much smaller than the cost of one that is too short. If you are unsure, wait longer.</p><p>That asymmetry is the reason I am not tearing out my scheduler. An imperfect expanding schedule errs in the forgiving direction, and any schedule beats reviewing everything every day. But it is a design decision, not a scientific result, and the difference is worth being honest about.</p><h3>The limit that matters most for language learners</h3><p>Here is the caveat I would put on the box. Nakata and Elgort found in 2021 that spacing reliably improves explicit vocabulary knowledge, the kind you can consciously retrieve, but not tacit knowledge, the automatic kind that shows up in fluent speech. A 2026 replication found neither schedule helped implicit knowledge.</p><p>This matches what learners keep reporting: a deck of two thousand mature cards and a mouth that still will not produce a sentence. Recognising a word on a card and reaching for it mid-conversation are different skills on different timelines.</p><p>It is not an argument against flashcards. Other work by Elgort found deliberately learned words do become genuinely integrated into the mental lexicon, and research on the testing effect suggests vocabulary translation is exactly the sort of simple, low-interference material where it works best. Flashcards are well chosen for vocabulary specifically.</p><p>But they build a substrate rather than a skill. Retrieval practice makes the word available; only use makes it automatic. This is why my own system keeps two separate schedules for every saved word, one for recognising it and one for producing it in writing, on the principle that recognising a word does not imply being able to produce it. That is a partial answer at best. The full answer involves talking to people.</p><h3>What to actually do</h3><p>Space your reviews rather than cramming them. This is the finding with the most evidence behind it and the one people most often skip.</p><p>Retrieve, do not re-read. If a card lets you look at the answer and think “yes, I knew that”, it is doing almost nothing. Produce the answer first, out loud or in writing.</p><p>Do not agonise over the algorithm. The scheduler is not the active ingredient, and no app has earned the confidence its interval curve projects. Any consistent spacing beats none.</p><p>When unsure, wait longer. Too-long gaps cost less than too-short ones.</p><p>Spend the words. A reviewed word is available, not automatic. The conversion happens in conversation, writing and reading, not in the deck.</p><p>The deck is a supply line, not the battle.</p><p><em>I build </em><a href="https://ownslides.com/"><em>OwnSlides</em></a><em>, which uses the SM-2 scheduler described above, with the caveats described above.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=0317bd594579" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Eleven Languages My App Can Describe But Not Pronounce]]></title>
            <link>https://medium.com/@ownslides/the-eleven-languages-my-app-can-describe-but-not-pronounce-2b7f4711e203?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/2b7f4711e203</guid>
            <category><![CDATA[language]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[language-learning]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Sun, 19 Jul 2026 13:29:32 GMT</pubDate>
            <atom:updated>2026-09-04T15:44:43.191Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Kurdish road signs showing place names in Arabic script with Latin romanisation below each: Akre, Soran, Bekhal, Harir." src="https://cdn-images-1.medium.com/max/1024/0*wiTSdcM8l3-u9xq3" /><figcaption>Kurdish signage: Arabic script above, Latin romanisation below. Kurdish is first on the list of eleven languages my app cannot pronounce. Photo by <a href="https://unsplash.com/@levimeirclancy?utm_source=medium&amp;utm_medium=referral">Levi Meir Clancy</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Most language apps offer thirty to forty courses, and mostly the same thirty to forty, because a hand-built course needs curriculum designers, native recordings and years of maintenance, and that only pays back at the top of the market.</p><p>The usual claim is that AI removes this constraint. I built on that assumption. My platform supports 91 languages today: Somali, Luxembourgish, Uzbek, Georgian, Xhosa, Hawaiian, Inuktitut, Esperanto among them. A year ago it was about seventy. There’s a comment in the codebase that still refers to “the 70+ supported languages” and to what happens when you add “language #71”, which I never updated and which now reads as a timestamp.</p><p>That drift is the actual subject here. The constraint on supporting a language moves, and it has moved twice already in the time I’ve been building this, and where it sits right now is not where anyone expects.</p><h3>The frontier is no longer knowledge</h3><p>The backend stores a phonological description of every supported language: three to six phrases about what makes it sound the way it does, fed to the model when it generates lessons.</p><p>Xhosa:</p><blockquote>Xhosa Latin alphabet — famous for CLICK consonants written c (dental), x (lateral), q (palatal); each click combines with voicing/aspiration/nasalization (e.g. gc, nq, ngx); Xhosa is TONAL (high/low) though tone is not written</blockquote><p>Hausa:</p><blockquote>Hausa Latin (Boko) alphabet including the hooked letters ɓ, ɗ, ƙ and the apostrophe ’y; ɓ, ɗ are implosive consonants; ƙ and ’ are ejective/glottalized; Hausa is TONAL (high, low, falling) and has phonemic vowel length; tone and length are usually unmarked in everyday Boko spelling</blockquote><p>Igbo is tonal with downstep. Zulu has breathy-voiced consonants that lower the tone of the syllable. Samoan has formal and colloquial registers where t↔k and n↔ŋ swap. Hawaiian distinguishes pau, paʻu and paʻū.</p><p>These were generated and then verified, not commissioned from a linguist. Five years ago this was the hard part of supporting a language, and the reason course catalogues stopped at forty. It is now close to free.</p><p>So the constraint moved somewhere else.</p><h3>The frontier is audio, for now</h3><p>I use two text-to-speech providers, covering 69 and 64 languages. Eleven of my 91 have a voice from neither: Kurdish, Yoruba, Igbo, Xhosa, Tajik, Turkmen, Scottish Gaelic, Esperanto, Māori, Hawaiian and Samoan.</p><p>Yoruba, Igbo and Xhosa have three of the most detailed phonological descriptions in the codebase and no audio at all. The system can explain that Xhosa has three click consonants and unwritten tone. It cannot pronounce one. Whatever “AI supports this language” means, it currently means something closer to can describe than can speak. Audio is the part that cannot be substituted, because pronunciation is learned by ear.</p><p>That list is a snapshot and I expect it to shrink, which is roughly how the last two constraints went. Eight of my languages were added in a single batch last June on the strength of one provider adding coverage the other lacked. The code assumes this: the record of which languages a vendor rejects is built at runtime from actual failures rather than hardcoded, specifically so that coverage can widen without anyone editing a config file. The eleven are not a verdict on those languages. They are where the edge happens to be this year.</p><p>There is a weaker version of the same problem, and it is the one that will bite first. Twenty-four languages have a voice from one provider and no fallback from the other. Luxembourgish is one. Coverage can narrow as well as widen, and when it does I hear about it from a user rather than from a changelog.</p><p>For a learner the useful move is to check for text-to-speech in your target language before building a routine around any app, and to re-check occasionally rather than concluding it is impossible. Where it genuinely does not exist yet, native recordings are the scarce input worth collecting deliberately: Wikipedia’s spoken articles, news radio, church services, diaspora YouTube.</p><h3>Before either of those, the frontier was rendering</h3><p>I expected displaying text to be solved. Each of these shipped to production before I noticed.</p><p>The ✓ and ✗ in grammar comparison slides rendered as empty rectangles in exported images. They sit in the Dingbats Unicode block, which the standard fonts do not cover.</p><p>The fix then failed in production while working locally. The font package registered the family as “Noto Sans Symbols Two” and I had written “Noto Sans Symbols 2”. My browser had substituted something close enough and shown me correct checkmarks; the headless renderer substituted nothing. A development machine has hundreds of fonts installed and will cover for a missing glyph rather than show you a box, so local rendering is not evidence.</p><p>IPA characters, meaning ɪ, ʃ, ʊ and ŋ, rendered as boxes under Korean display fonts, which do not carry the phonetic blocks. A pronunciation feature displaying boxes instead of phonetic symbols is worse than no pronunciation feature.</p><p>CEFR badges started rendering “A1” in the wrong typeface, because the colour emoji font was winning the cascade for plain ASCII digits. The fix is a hand-written list of 28 Unicode ranges limiting that font to actual emoji; there was no clean range that included the emoji and excluded the numbers.</p><p>Chinese, Japanese and Korean share Unicode codepoints for characters drawn differently in each language. A single cross-script font chain renders Japanese text with Chinese glyph forms, wrong to a native reader and invisible to me. The fix is per-locale font loading.</p><p>Loading those fonts properly then broke the app. Eight families across five weights and several subsets emitted around 457 font preload tags per page. iOS Safari holds first paint until they arrive, which put cold launch of the installed app on a Korean mobile connection at roughly ten seconds. The CJK font CSS alone was about 1.55MB uncompressed and render-blocking, close to three quarters of the landing page’s blocking CSS.</p><p>Glyph coverage, font cascade, regional glyph forms and load performance each failed separately. Support for a writing system is not one property a tool either has or lacks. This class of problem is mostly solved now, in the sense that the fixes exist and I have applied them, which is what a frontier looks like from the far side.</p><p>For a learner: paste a paragraph of your target language into any tool before committing to it. Boxes mean missing glyphs. Harder to catch, and worth asking a native speaker about, is text rendering in a font that has the characters but was never designed for the script. That is how you spend a year learning slightly wrong letterforms.</p><h3>The constraint that does not move on its own</h3><p>Vendors add voices and font stacks improve without my involvement. Assumptions in my own code do not.</p><p>I debugged the lesson-generation pipeline against Lithuanian: non-trivial script, heavy inflection, not English. The quality gates written during that period encoded three assumptions that do not generalise. That spelling and pronunciation always diverge, true in Lithuanian and Russian, false in English, where “ough” covers five pronunciations. That a vocabulary item is a single word, false for English, where “you know what I mean?” is a unit worth teaching. That the target language is foreign to the learner, false for English taught in English, where translation exercises have nothing to translate between.</p><p>Each assumption made the gates reject valid lessons, and I fixed them one at a time with checks for English, which is how language support becomes thirty unreviewable conditionals. The replacement is a per-language profile: gates ask whether a language distinguishes spelling from sound rather than whether it is English. Four properties, listed explicitly per language, so that adding a language means writing its profile instead of auditing every gate. That refactor is the only reason going from seventy languages to ninety-one was tractable.</p><p>The clearest case came from a gate meant to catch the model emitting an alphabet chart in place of vocabulary. It treated any token of one or two characters as a letter, which holds for alphabetic scripts and not for Korean, where a single character is a whole word.</p><blockquote>한 개 (one item) and 네 잔 (four glasses) are two single-syllable words, not an alphabet row.</blockquote><p>The gate rejected every Korean numbers batch in production until I found it. The failure was not in the model; it was in my definition of a word.</p><h3>What this means for studying a less-common language</h3><p>Check rendering first. Paste real text into every tool you plan to use and look at it.</p><p>Check whether the language has synthetic speech, and check again in six months if it does not. The answer changes.</p><p>Use AI as a drill partner rather than an authority. For the top few hundred languages the models are useful, and they told me things about Hausa consonants I then verified. They are also wrong in ways a beginner cannot detect. My own pipeline produced three contradictory pronunciations for the same Lithuanian word on one page before I constrained it. Anchor to native audio.</p><p>Build vocabulary from frequency rather than textbook chapter order, since small-language textbooks tend to be old and formal, and put the words into spaced repetition immediately.</p><p>Use media as the syllabus when no course exists: music, news, TikTok, church services. A page or a minute a day, mined for words, is a workable intermediate method.</p><p>Learn the script from the start. Romanisation is faster for two weeks and a ceiling afterwards, and the standards are often lossy. Thai’s official RTGS romanisation marks no tones and drops vowel length, so studying Thai through it means studying Thai with the tones removed.</p><h3>One unadvertised benefit</h3><p>Studying a rarely-learned language changes how speakers respond to you. People whose language is widely studied are politely encouraging; people whose language is rarely studied tend to be delighted, and conversations become events. It is a motivational advantage that Spanish learners do not get.</p><p>It is also why I keep working at this end of the list, where the vendor coverage is thin and the edge cases are real. The count was seventy, then ninety-one, and the interesting work is wherever the constraint has moved to next.</p><p><em>I build </em><a href="https://ownslides.com/"><em>OwnSlides</em></a><em>, which supports 91 languages as of writing, including eleven it can describe in detail and cannot yet pronounce.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2b7f4711e203" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[30 Post Ideas for Language Teachers Who Ran Out of Content]]></title>
            <link>https://medium.com/@ownslides/30-post-ideas-for-language-teachers-who-ran-out-of-content-8c5c04fa2201?source=rss-419a7c8bdef8------2</link>
            <guid isPermaLink="false">https://medium.com/p/8c5c04fa2201</guid>
            <category><![CDATA[teaching-and-learning]]></category>
            <category><![CDATA[language-learning]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[social-media-marketing]]></category>
            <dc:creator><![CDATA[OwnSlides]]></dc:creator>
            <pubDate>Sun, 19 Jul 2026 12:01:01 GMT</pubDate>
            <atom:updated>2026-07-19T12:01:01.932Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A working list to steal from, organized by what each idea is good at: growth, saves, or comments.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*TmQe5fqg9xomaeRN" /><figcaption>Photo by <a href="https://unsplash.com/@codzilla_swiss?utm_source=medium&amp;utm_medium=referral">AbsolutVision</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Every teaching account hits the wall around week six. You have posted the obvious vocabulary lists, the alphabet, the greetings, and now the queue is empty and the feed is hungry. This list is what I reach for when that happens. I run a language-learning content platform, so producing these formats is literally my job; the disclosure is that my team builds a tool (OwnSlides) with templates for most of them, but every idea below works with any tool.</p><p>The ideas are grouped by the job they do. A healthy feed mixes all three.</p><h3>Growth posts (reach new people)</h3><ol><li><strong>Words you already know.</strong> Loanwords from English in your target language. Instantly encouraging for beginners, extremely shareable.</li><li><strong>False friends.</strong> Words that look the same and mean something embarrassing. The embarrassment is the engagement.</li><li><strong>This word has no translation.</strong> Untranslatable words with a story. The classic for a reason.</li><li><strong>Country facts through language.</strong> Why a language has seven words for rice, or none for “please”.</li><li><strong>Did-you-know statistics.</strong> “Korean is spoken by 82 million people, and its alphabet was invented in one project.” Numbers stop the scroll.</li><li><strong>The idiom, literally.</strong> Draw or describe the literal image of an idiom, then reveal the meaning.</li><li><strong>Slang your textbook refuses to teach you.</strong> Keep it clean enough for the platform, spicy enough to share.</li><li><strong>How natives actually say it.</strong> Textbook phrase vs street phrase, side by side.</li><li><strong>The hardest word to pronounce.</strong> Dare people to try in the comments or in a voice note.</li><li><strong>Same word, three countries.</strong> Regional differences (Spanish of Spain vs Mexico vs Argentina).</li></ol><h3>Save posts (become someone’s study material)</h3><ol><li><strong>Word of the day, but themed weeks.</strong> Monday to Friday all food, next week all travel. Themes make people come back.</li><li><strong>The 10-word starter pack.</strong> For a situation: airport, pharmacy, first date.</li><li><strong>Conjugation table of one high-value verb.</strong> One verb, all the forms people actually use, nothing else.</li><li><strong>Grammar rule + three examples.</strong> One rule per post. Three examples, one slightly funny.</li><li><strong>A vs B comparison.</strong> Por vs para, ì vs ë, ser vs estar. The eternal confusions are eternal traffic.</li><li><strong>Sentence patterns.</strong> “If I had ___, I would ___.” One pattern, five fillings.</li><li><strong>A tiny dialogue.</strong> Four lines, real situation, natural replies. People screenshot these.</li><li><strong>Numbers cheat sheet.</strong> Prices, dates, phone numbers, the stuff people actually fumble.</li><li><strong>Pronunciation trap of the week.</strong> One sound, minimal pairs, how to place your tongue.</li><li><strong>Reading passage, A2 level.</strong> Three sentences with a glossary. Rare on social, hugely appreciated.</li></ol><h3>Comment posts (make the algorithm notice)</h3><ol><li><strong>Fill in the blank.</strong> Leave the blank in the caption too, people answer without leaving the feed.</li><li><strong>Spot the mistake.</strong> One wrong sentence. Teachers cannot resist correcting it, learners neither.</li><li><strong>Order the sentence.</strong> Scrambled words, ask for the correct order in comments.</li><li><strong>Which one is correct?</strong> Two options, both plausible. The 50/50 gets the most replies.</li><li><strong>Translate this if you dare.</strong> One genuinely tricky sentence.</li><li><strong>What would you say here?</strong> A situation, ask for responses in the target language.</li><li><strong>Caption this in [language].</strong> A photo, answers in comments, you correct the best ones.</li><li><strong>Poll the pain.</strong> “Which grammar topic broke you?” Then teach the winner next week, which is also a free content plan.</li><li><strong>Guess the meaning from context.</strong> One hard word inside an easy sentence.</li><li><strong>The weekly quiz recap.</strong> Five questions from your own last five posts. Rewards followers, punishes nobody.</li></ol><h3>Making this sustainable</h3><p>The trap with a list like this is doing each idea by hand, from scratch, forever. The way out is templates: each idea above is really a data shape (a word list, a rule with examples, a dialogue, a pair to compare), and if your tooling knows the shape, next week’s post is just new data in the same clothes. That is the entire thesis of OwnSlides: you pick the content type, the AI drafts the deck, you correct it, and it exports as a carousel or a short video. One idea from this list, produced weekly with a template, is a series. Ten of them is a content calendar.</p><h3>FAQ</h3><p><strong>How do I pick which ideas to use?</strong> Pick two growth, two save, one comment format, and run them as weekly series for a month. Series beat one-offs because followers learn what to expect.</p><p><strong>Should every post be in both languages?</strong> Hook in the learner’s language, content in both. A post entirely in the target language filters out the beginners who make up most of your audience.</p><p><strong>What if my niche language has a tiny audience?</strong> Tiny audiences are loyal audiences. Ideas 1, 3 and 5 also reach the curious general public, which is where new learners come from.</p><p>The tool with templates for most of these is at <a href="https://ownslides.com/">ownslides.com</a> if you want to try the workflow. And if you want a fun way to keep your own streak going, the free daily games are a good habit anchor.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8c5c04fa2201" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>