A scientific article is like a signature. Every country has its own research handwriting. Over 30 years, three journals accumulated 5281 such 'autographs' from 81 countries.

To decipher this handwriting, Chengzhi Zhang's team created a program that learns from examples (that's machine learning). The program reads an article and picks out snippets where the method is described: for example, surveys or citation counting. It understands the natural language of science and decides what's important on its own. At its core is the ability to focus on what matters (attention mechanism) and break text into parts (transformers). The accuracy of recognizing 15 methods is 0,83 by a special measure.

It turns out that countries have different habits. In Belgium and the Netherlands, nearly half of the works simply count how many times an article was cited. In China and Brazil, they run experiments. To measure diversity, the scientists applied a formula that ecologists use to count species. This number comes from information theory: H = -∑ $p_i$ log₂ $p_i$. Here $p_i$ is the proportion of articles using each method. The higher H, the more diverse the approaches.

Over the years, national handwritings converge. Scientists increasingly work together: the average number of countries per article rose from 1,1 to 1,45. Differences between countries are fading, confirmed by a test from probability theory. The share of articles with multiple methods grew from 15% to nearly 50%. To see the big picture, the data was compressed into a two-dimensional map (dimensionality reduction) and three groups of countries emerged: 'mature' (USA, UK), 'fast-growing' (China, Australia), and 'traditionalists' (Belgium).

Such a map of scientific handwritings helps select partners. If you need an experimenter, look for a Chinese colleague; for bibliometric analysis, head to Belgium. In the future, programs will be able to learn without hints (self-supervised learning) and use Large Language Models to not only recognize methods but also explain their choices. Perhaps we will see the handwriting of science become common, like an inscription made by many hands.