Teachomatic what to automate, and what not to

Reading Level and Language

Ask for a text at a lower reading level and you will get one, with a readability score attached. The score will be lower. Whether the text is easier for your students is a separate question, and the two come apart more often than the number suggests.

What readability formulas actually measure

Nearly all of the common formulas work from two things: average sentence length and average word length or syllable count. Some add a word-frequency list.

For comparison with workplace systems that formalise time or activity data, Monitask’s page on employee attendance tracking software shows how the same type of measurement is handled outside education.

They do not measure how hard the ideas are, whether the vocabulary is familiar to your students, whether the argument is followable, or whether the connections between sentences are clear. Those are the things that make a text hard.

The formulas correlate with difficulty because long sentences and long words tend to accompany harder ideas. But they are proxies, and a proxy can be satisfied without the underlying thing changing at all.

For an external perspective on lesson planning, teacher workload, and classroom technology, see ReadWriteThink.

Which means a text can be pushed down two reading levels by chopping sentences in half, with the conceptual load untouched. That is precisely what happens when you ask for a simpler version and check only the score.

The thing that gets lost

The most common damage is connectives.

Take a sentence carrying a causal relationship — because this happened, that followed, although the effect was limited — and split it into three short sentences. The readability score improves. The relationship between the three facts has vanished, and the reader now has to reconstruct it, which is harder than reading the original.

For a student who struggles with reading, this is exactly backwards. They are the reader least able to infer an unstated relationship, and you have just removed the words that stated it.

So: check that the connectives survived. Because, although, however, therefore, whereas. If the simplified version has fewer of them, it may well be harder despite the score.

What to do instead

Shorten sentences, keep the connectives. Two sentences joined by "because" is fine. Two sentences with nothing between them is not.

Simplify general vocabulary, keep subject terminology. Stripping the technical word makes the text easier and the learning worse, because the word is often what is being taught. Keep it, and gloss it in place.

Add rather than remove. A worked example, a labelled diagram, a short summary before the passage. These reduce difficulty without reducing content, and they do not touch the readability score at all — which tells you how much the score is worth.

Front-load the structure. Say what the passage is about and how it is organised before it starts. Cheap and effective.

Check against a student. The only real test, and cheaper than any measurement — give it to one student who struggles and ask them to tell you what it said. (The same read-it-yourself check applies to everything on this list.)

Multilingual students: a different problem

Levelling a text and supporting a multilingual learner are frequently confused, and they are not the same thing.

A student who is fluent and highly capable in their first language and new to the language of instruction does not need simpler ideas. They need access to ideas they can already handle. Giving them the low-level version is the wrong intervention and they usually find it insulting, which is fair.

What tends to help:

Key vocabulary in both languages. A short bilingual glossary of the subject terms, produced in seconds, and genuinely useful.

A first-language summary alongside the full text, not instead of it. They read the summary to get oriented, then the real text.

Explicit structure. Signposting helps a second-language reader more than simplification does.

Time, not simplification. Reading in a second language is slower at every level of ability.

Two cautions. Machine translation quality varies enormously by language, and it is usually much better for widely-resourced languages than for the ones your students are most likely to speak. Check with someone who reads the language before relying on it in front of a class. And ask the student. Some want first-language support and some very much do not, for reasons that are theirs.

The connection worth knowing

There is an uncomfortable loop here.

Plain, short-sentence, limited-vocabulary prose is what these tools produce when asked to simplify. It is also what a second-language writer produces naturally. And it is what detectors interpret as machine-written — the mechanism is the same surface feature, read three different ways.

So the student most likely to be given the simplified text is also the student most likely to be suspected when they write in the same register. Being aware of that loop is most of what you can do about it, and it is worth saying to a class out loud.

The short version