When a child spells a word wrong, the mistake is never random. It lands on one of three layers of how English spelling actually works: the sound of the word (phonology), the way that sound is turned into letters (grapheme-phoneme mapping), or the way the word is built from parts (morphology). Linguists and speech-and-language specialists have used those three layers for decades to explain why a speller struggles - not just that they did.
spelling.live's pattern engine is built to span all three. When it recognises one of the 18 KS2 spelling patterns behind a mistake, that pattern always belongs to at least one of these layers - and often sits across two, because real spelling mistakes rarely respect a tidy boundary. This post walks through each layer, shows which of the app's live patterns fall into it, and explains why covering all three matters more than any single one.
Why three layers, and not just "phonics"
Ask most people how children learn to spell and they'll say one word: phonics. Phonics is essential - but it's only the first of the three things English asks of a speller. Sounding a word out gets you to frend for friend, foto for photo, hopping spelled like hoping. Every one of those is a perfectly reasonable guess by ear. What English then demands is that you know which of several possible spellings is the right one, and how the word changes shape when you add a prefix, a suffix or make it plural.
So a spelling system that only understands sound can tell you a child spelled photo as foto - but it can't tell you whether the child couldn't hear the word, chose the wrong letters for a sound they heard perfectly well, or lost the root when building a longer word. Those are three different problems with three different fixes. Naming the layer is what turns "wrong" into "here's what to practise".
The short version: phonology is the sound, grapheme-phoneme mapping is choosing the letters for that sound, and morphology is building the word from meaningful parts. A strong speller has to get all three right; a diagnosis is only useful if it can tell them apart.
Layer 1 - Phonology: hearing the word
Phonology is about the sounds in a word, before any letters are involved. A phonological mistake is one where the child's ear leads them astray: an unstressed vowel that hides which letter is really there, or a word spelled exactly as it sounds.
| Pattern in the app | Example | The sound problem |
|---|---|---|
| Unstressed vowel (schwa) mistakes | separate → seperate | The middle "uh" sound gives no clue that the letter is an a. |
| Phonetic spelling | definitely → definitly | Spelled precisely by ear, dropping a syllable the mouth glosses over. |
| Long vowel patterns | cake → cak | The vowel says its own name, but the sound alone doesn't say how to write it. |
The schwa is the classic phonological trap, and it's the one the engine works hardest to get right. An unstressed "uh" appears in chocolate, definite, family and hundreds of other words, and by ear it could be almost any vowel. Rather than assume every vowel slip in a hard word is a schwa mistake, the engine checks each word against real speech-stress data, so it only flags a vowel as unstressed when that's genuinely true. That's the difference between naming the sound problem correctly and guessing.
This is also the layer that connects to how the app teaches - the phonics audio and phoneme breakdowns that let a child hear "sh-o-p" broken into its sounds. Diagnosis and instruction meet on the phonology layer.
Layer 2 - Grapheme-phoneme mapping: choosing the right letters
This is the layer people most often mean when they say English spelling is "irregular". A grapheme is a letter or letter-group; a phoneme is a single sound. Grapheme-phoneme correspondence (GPC) is the mapping between them - and in English it is famously many-to-many. The /f/ sound can be written f, ff or ph. The /s/ sound can be s, ss or c. A child can hear a word perfectly and still pick the wrong grapheme for a sound.
The app has a pattern named for exactly this - "Same sound, different spelling" (gpc_error under the hood) - alongside several patterns that are really specific GPC choices:
| Pattern in the app | Example | The letter-choice problem |
|---|---|---|
| Same sound, different spelling | photo → foto | A plausible but wrong grapheme for the /f/ sound. |
| Vowel digraph confusion | believe → beleive | The "i before e" pair - two letters, one sound, easy to flip. |
| Silent letters | knife → nife | Letters written but never sounded (kn-, wr-, mb, the t in castle). |
| Consonant pattern mistakes | watch → wach | The ch/tch choice, or soft c and g. |
| Double-letter mistakes | necessary → neccessary | Which consonant doubles - and which doesn't. |
| Foreign-origin spelling | tongue → tong | A que/gue ending borrowed from French. |
Under the hood, this layer is powered by a dedicated table of grapheme confusions - ph→f, wh→w, kn→n, wr→r, ght→t, the r-controlled vowels er/ir/ur, the que→k and gue→g endings. Each confusion is tied to the pattern it best explains, so a mistake like foto is named as a sound-to-letter choice, not lumped in with a random letter swap. Silent letters are the sharpest example: the sound gives the child no signal that the letter is there at all, which is why they rank among the hardest patterns to master.
Layer 3 - Morphology: building the word from parts
Morphology is about meaningful units - roots, prefixes and suffixes - and how they join up. A morphological mistake isn't about sound or letter choice; it's about the structure of the word. The child who writes unecessary for unnecessary heard the word fine and knows the letters; they just lost the doubled letter where the prefix un- meets the root necessary.
| Pattern in the app | Example | The structure problem |
|---|---|---|
| Prefix mistakes | unnecessary → unecessary | A prefix joins the root and a letter is lost at the seam. |
| Suffix mistakes | station → stashun | An ending like -tion written by sound rather than by form. |
| Root word mistakes | medicine → medisin | The shared root shifts across a word family (medical → medicine). |
| Word family mistakes | signature → signiture | Losing the root that binds sign → signal → signature. |
| Plural mistakes | children → childs | Forming more than one - regular or irregular. |
| Short vs long vowel doubling | hopping → hoping | Whether the consonant doubles when a suffix is added. |
This is the layer that phonics alone can never reach, and it's where older, fluent spellers do most of their remaining work. The engine detects it structurally - checking the boundaries where prefixes and suffixes attach, recognising the assimilated Latin prefixes (il-, im-, ir-), and telling apart a short-vowel root that doubles its consonant (hop → hopping) from a long-vowel one that doesn't (hope → hoping). Seeing that signiture is a root mistake, not a random misspelling, points a parent straight at the fix: teach the sign family together, and the whole cluster gets easier at once.
Spot the layer: five real mistakes
The best way to feel the difference is to try it. Here are five mistakes exactly as a child might write them. Before you read the answer, ask yourself: was the trouble in the sound, the letters chosen for that sound, or the way the word is built? These are the same five judgements the pattern engine makes automatically.
1. seperate for separate. The middle vowel is an unstressed "uh" - by ear it could be an a or an e. The sound itself gives the game away, so this is a phonology mistake (the schwa). Fix: say it as "sep-a-rate", stretching the hidden vowel.
2. foto for photo. The child heard the /f/ perfectly - they just picked f instead of the ph that this word wants. Right sound, wrong letters: a grapheme-phoneme mistake. Fix: practise the ph = /f/ family together (photo, phone, dolphin, elephant).
3. unecessary for unnecessary. Sound and letters are both fine - the doubled n was lost at the seam where the prefix un- joins the root necessary. A morphology mistake. Fix: build it in two visible parts, un + necessary, and the double n stops disappearing.
4. hear for here. A real, correctly-spelled word - just the wrong one. It's both phonology (the two words sound identical) and grapheme-phoneme (the letters spell a different meaning). The engine flags it as a homophone and names both. Fix: anchor the meaning - "you hear with your ear".
5. hoping for hopping. The child needed to double the p before -ing because hop has a short vowel - hope wouldn't. That doubling rule lives at the join between root and suffix, so it's a morphology mistake (with an orthographic edge). Fix: contrast the pair directly - hop → hopping vs hope → hoping.
Notice that three of the five were not phonics problems at all - and a phonics-only tool would have mislabelled every one of them. That is the whole case for naming the layer.
The layers overlap - and that's the point
Real mistakes don't always sit on one layer. A homophone mix-up - writing hear for here - is both a phonology problem (the two words sound identical) and a grapheme-phoneme one (the child chose a real, correctly-spelled word with the wrong letters for the meaning). The app treats it as both, rather than forcing it into a single box.
That honesty about overlap runs through the whole engine. When a mistake genuinely matches more than one layer, the classification reflects it. And when a mistake is a jumble that doesn't reduce to a single clear cause, the engine says so rather than inventing a confident-sounding label - because a wrong diagnosis is worse than an honest "unclear". A pattern report is only trustworthy if it knows the limits of what it can tell.
Why covering all three matters
A spelling tool that only understands one layer will keep misreading the other two. It'll call a morphology mistake a "typo", or a grapheme choice a "sounding-out error", and send a child to practise the wrong thing. Covering sound, letter-choice and structure together is what lets the app say not just what word was missed, but which kind of knowledge was missing - and therefore what to practise next.
That's the same principle behind the app's Skills Check and its spaced-repetition scheduler: the more precisely we know the layer a child is struggling on, the less time they waste on the two layers they've already got. A child who keeps tripping on the schwa doesn't need more silent-letter drills. A child losing roots across word families doesn't need more phonics. Name the layer, and practice stops being random.
Spelling isn't one skill - it's three, stacked on top of each other. Once you can see which layer a mistake belongs to, the next word that shares it gets a whole lot easier.