The science behind Spelling Garden

How Spelling Garden teaches spelling, the research each part of it draws on, and which parts are our own judgement rather than established findings.

1. Recall, not recognition: hear it, then type it

Every spelling question is dictation. You hear the word, see a sentence with a gap where it goes, hear the sentence, and type the word from memory. Nothing is ever put in front of you to copy, and there is no mode where you simply read through a list.

Pulling an answer out of memory strengthens it more than looking at it again. This is the testing effect: Roediger & Karpicke (2006) found that taking a test on material improved how much people remembered later, more than spending the time restudying it. Producing the answer yourself helps too (the generation effect, Slamecka & Graf, 1978), and for spelling, typing the word is exactly the skill you want to use afterwards.

2. Starting at your level

Your first visit opens with a short placement test. It starts with fairly easy words and adjusts as you go: spell one right and the next is harder, get one wrong and the next is easier, and each step gets smaller whenever the direction changes, so it quickly settles on the level where you start to make mistakes. It ends once it has found five words you can't spell (or after 50 words), and those five become the first plants in your garden.

This is an adaptive staircase, a method used for decades in perception research to find a threshold without testing every level (Levitt, 1971). Words that sound exactly like another word, such as champaign and champagne, are left out of the test, because dictation alone can't tell them apart.

3. Only the words you get wrong

After the test, each day begins by digging up new words at and just above your level. A word you spell correctly first time is treated as known and is never asked again. A word you get wrong becomes a seedling that you plant in your garden. If you press "I don't know this word", it is planted as a mushroom, so you learn what it means as well as how it is spelled. Digging stops once five new words have turned up.

So your garden holds only words you have shown you personally can't spell, rather than a fixed list everyone works through. Getting a word wrong first is not wasted effort: trying to answer and then being shown the right answer leads to better learning than just being shown it, even when the attempt fails (Kornell, Hays & Bjork, 2009), and errors followed by correction are a strong basis for learning (Metcalfe, 2017).

4. Spaced repetition: gaps that grow

A newly planted word comes back the next day. Spell it right then and it waits three days. After that, every time you spell it right the gap grows about two and a half times, from days to weeks to months. Get it wrong and the gap drops to a fifth of what it was, and never below a day. A plant that is due wilts until you water it.

Spreading practice over days gives far better long-term memory than the same amount of practice in one sitting, a finding that goes back to Ebbinghaus (1885) and holds up across a large body of research (Cepeda et al., 2006). The best gap grows with how long you want to remember something (Cepeda et al., 2008). Multiplying the gap after each success and cutting it after a failure follows the SuperMemo SM-2 algorithm (Wozniak, 1990), which Anki also grew from.

5. Only your first try each day moves the schedule

Get a word wrong while watering and it goes to the back of the queue, and you spell it again before you finish. But only your first attempt of the day decides when the word comes back. Getting it right a minute after being shown the answer doesn't count as knowing it. Extra practice after the day is done works the same way: it is there for anyone who wants more, and it changes nothing about when your words come back.

Repeating something moments after seeing it feels like learning because the answer is still fresh, but that fluency fades fast. Bjork & Bjork (2011) explain why conditions that make practice feel harder, such as spacing it out, tend to produce more durable learning, and Karpicke & Roediger (2008) found that once something has been recalled, studying it again does little for long-term memory while retrieving it again later does a lot. Tomorrow's first attempt is the honest test.

6. Seeing exactly where you went wrong

When you misspell a word you see your spelling above the correct one, with the part that differs marked on both: seperate against separate. The mark covers one unbroken stretch rather than a scatter of single letters, so a jumbled ending reads as "this part is wrong" rather than a puzzle.

Feedback that points to the specific part of a task that went wrong is among the most effective kinds (Hattie & Timperley, 2007), and correcting an error straight after it is made is where much of the benefit of making it comes from (Metcalfe, 2017).

7. Knowing what a word means, not just how it looks

When a word you didn't recognise comes due, it is asked in two parts. First you hear it, without seeing it written, and choose its meaning from three definitions: one real and two wrong. Then you spell it. It only counts as known when you get both right, because spelling a word you can't define isn't really knowing it. The wrong definitions change from one review to the next, so the answer can't be remembered by its position or its look, and they are written to be plausible rather than silly.

Strong readers and spellers hold words as high-quality representations in which spelling, sound and meaning are tightly bound together (Perfetti & Hart, 2002; Perfetti, 2007). Multiple-choice questions can help learning when the wrong options are plausible enough to need real thought (Little et al., 2012), which is why ours are written with care and checked so the real answer never gives itself away by its length.

8. Words in sentences, and sound-alikes together

Every question puts the word in a sentence, and each word has several sentences that take turns, so you meet it in different settings over time rather than learning one line by heart. Words that sound the same but are spelled differently, such as their, there and they're, are planted together as a family in a pond in your garden, and each sentence is written so that only one of them fits.

Words become easy to recall when their spelling is linked to their sound and meaning across many varied encounters (Ehri, 2014), and learning vocabulary in context is well established (Nation, 2001). For words that are easily confused, meeting them side by side helps you notice what tells them apart (Kang & Pashler, 2012).

9. Why a word is spelled that way, and your own memory hooks

Open any word in your garden to see where it comes from, how it breaks into parts (accommodate is ac- + com- + mod + -ate, which is where its double c and double m come from), each of its meanings, and how it is getting on in your schedule. You can also write your own memory hook for a word. It shows as a small tag on the plant, and only you can see it.

Understanding how a word is built, and what its parts mean, helps both spelling and vocabulary (Bowers, Kirby & Deacon, 2010; Bowers & Kirby, 2010). Memory cues people make for themselves tend to work better than cues they are given (Mäntylä, 1986), which is why the hooks are yours rather than ours.

10. A little every day, at a pace you can keep

The garden gives you one short session a day: dig up a few new words, plant them, then water everything that is due. Once that is done, the day is complete. New words are held back altogether while ten or more plants are already due, so a heavy week of reviews never has more piled on top of it, and digging resumes as words move to longer gaps.

Spacing study sessions out gives longer-lasting memory than bunching them together (Bahrick & Phelps, 1987). Behind the scenes, how hard each word really is gets measured from how people do the first time they meet it, the way chess ratings work: each result nudges both the word's difficulty and the player's level (Elo, 1978; Pelánek, 2016). That measurement is what decides which new words you are offered next.

Plenty of the specifics are sensible design choices rather than findings, and we would rather say so.

• Five new words a day. Enough to keep a garden growing without swamping the reviews. No study says five is the right number.
• Holding back new words once ten are due. Ten is about two days of new words. The number is ours, and we will watch how it plays.
• The exact gaps. Three days after the first success, then two and a half times longer each time, and a fifth after a mistake. These follow the shape of well-tested scheduling algorithms, but the values are a starting point that we will tune as we learn from real play.
• One right answer first time means you know a word. A single correct spelling can be luck. We accept that some of those words may catch you out another day, so that you are never drilled on words you already know.
• The placement test's settings. Where it starts, how big its steps are and when it stops were tuned by hand.
• Three options for meaning. Two wrong definitions balance a fair question against a quick one.
• British and American spellings. Colour and color are both correct, so both are marked right. That is fairness, not research.
• The garden itself. Seeing progress build up is motivating (Locke & Latham, 2002), but plants, ponds, wilting and the rest are our design.

• No multiple-choice spelling. Every spelling is typed from memory. The only multiple-choice question is about meaning, and it is always followed by typing the word.
• No credit for cramming. Second attempts and extra practice never move a word's schedule. Only your first try on a watering day does.
• No flood of new words. Five a day at most, and none while your garden already has plenty to water.
• No one-size-fits-all list. Words you already spell correctly are left alone.
• No punishment for a missed day. Overdue plants wilt, but they are still there, and they pick up where they left off when you water them.