How we know

Parauling's guides explain what scientific research says about how children learn words. This page explains how we write them: what we read, how we decide how certain each point is and how we check that every sentence says the same thing as the article it comes from.

In short

• First we read the research reviews, then the studies with children. Always the full text, not just the abstract.
• Every claim we make comes with two answers: how well established it is in general, and how many studies there are with children learning a language.
• Every sentence that makes a claim is checked again, against the original text, by an AI agent that did not write it. At the end of each guide you will find the sources, with a link to each article.

What we read

A study is an experiment: for example, one group of children learns some words one way, another group learns them another way, and a few days later the researchers check how many each group remembers. A review is an article that brings together all the studies done on the same question and compares their results. A single study can give a result by chance; a review tells you what happens when the question has been looked at many times.

That is why we start with the reviews and with the work everyone in the field cites, and draw a map: what is taken as established, what is debated and what has not been studied. Only then do we move on to the studies with children, and we read them within that map. The order matters: if you start with the handful of studies on children, "little studied with children" can be read as "nobody knows if it works", and that is not the same thing.

We also look at two more things: what education bodies recommend (and where they get it from), and the claims that circulate online about the topic, which we trace back to the source they cite. That is where the "What people say that isn't quite right" section of our guides comes from.

How we say how certain each point is

Every claim answers two questions, and both are needed.

How well established is it, in general? Across all studies, at all ages. In the guides, the box under each heading tells you:

And with children learning a language? Many studies, some, few or none, and whether they say the same thing as the research in general. When there are few, we say how many: "three studies, from two teams". When there is a single study, we say how many children took part and how old they were.

Both are needed because they often don't match. Something can be well established in adults and have very few studies with children; that doesn't mean it doesn't work with children, it means it has hardly been studied with them. And when studies with children say something different from those with adults, we say so too.

We never write "scientifically proven". When something isn't known, we say so, and we say what the doubt is about: whether it works, how much or for whom.

How we choose our sources

• Two reviews that share authors, or many of their studies, count as one. Two confirmations from the same team are not two independent confirmations.
• We don't choose journals by their impact factor (a number that counts how often their articles are cited): choosing by that number would leave out leading journals and let dubious ones through. We check whether the journal is on the lists that assess journal quality (such as the national lists of Finland, Norway and Denmark) and who publishes it.
• Some journals publish almost anything, after a quick review of uneven quality. We don't rule them out, but an article from one of these journals never supports a claim on its own, and if it is the only source, the sentence says so.
• We check that no article we cite has been retracted or corrected in a way that changes its results.

How we check every sentence

• Each review and each study with children is read separately by two AI agents (programs that each do one specific task), without either seeing what the other has done, and they record its data with a verbatim quote and the page number. Where they disagree, a third goes back to the article and decides which is right.
• Every sentence that makes a claim is linked to its source, with the verbatim quote from the article and the page number.
• Another agent, one that did not write the sentence, checks it against the text of the article. It doesn't see how certain we said it was: only the sentence and the article, and its job is to look for errors. If it finds that the sentence says more than the article does, or a figure that isn't there, we go back to the article, rewrite the sentence and check it again.
• A different agent, one that has seen none of our work, reads the whole text and says what it understands from it: if it comes away thinking something is better established than it is, or less, the text is rewritten, even if every sentence is true.

Who does it

I'm Marina Mele. I decide which topics we review, which questions we ask and what goes into each guide, and I read every page or guide before it is published. My background is in research, in a different field: I have a PhD in quantum physics.

We do this work with a written, fourteen-step method, and with AI agents that find the articles, read them, extract their data and check the sentences, each with a narrowly defined task. The method is what keeps every sentence linked to a quote and a page in the article it comes from.

What we are not: we are not a research group, we don't run our own studies with children, and the guides are not reviewed by other scientists the way an article is before publication. They are a summary of published research, done carefully, and anyone can check it.

When a new study comes out, or if you find an error

Each guide shows the date it was published and, when it changes, the date of the latest update and what has changed. When a new study comes out, we go back over the sentences that depend on it and change them if needed. If a conclusion changes (for example, something we described as well established becomes debated), the guide explains the change; we don't quietly delete the old sentence.

If you find an error, write to us at support@parauling.com: we will check it against the article and, if needed, correct it and say so.