The Fractal Structure of Language

New research points to an underlying fractal structure of language. This means that using math, we can analyze language at different scales to define its distribution frequency, examine its underlying structure, and dissect word formation. There has always been linguistics (the study of language), but it is only in recent years that computers have been flexible and powerful enough to go into deeper analysis.

Internal language organization

What do we see? Recent studies from Google’s DeepMind research branch that measure language parameters show that language perhaps has an unspoken large-scale internal order, and a repeating structure. Furthermore, word choice is not independent (what you wrote paragraphs ago still influences what you write now, long past when it statistically should, or makes any logical sense), and it also has self-similarity (internal patterns that repeat when there is no reason for them to repeat). Spooky.

So, if we look at our results, we can compare and contrast what should be true vs. what really happens. If we have a random string of letters, we would logically expect them to have few words created in sequence. We might get a letter next to a vowel, but there might be useless nonsense next to the letter-vowel sequence. To illustrate, we might get C-A, but unless the next letter happens to be one of the handful that forms a real English syllable, the sequence breaks; the next letter will send us back to the beginning of our wait. This is just a sample; if we want a three- or four-letter word or even more, natural words will be few and far between.

But, of course, letters do not come randomly. They are naturally distributed in our words unevenly; E occurs more frequently than the letter W. Each letter exists in a weighted form statistically.

What the tool does

These tools were made to examine word creation from an English-mimicking string of letters. The database of letters follows English’s natural letter frequency array, so the words that do emerge reflect English’s built-in statistical structure rather than pure chance.

The first tool on this page keeps it simple; we look for three- or four-letter words. To ensure weighted randomness and analysis we send a request to the API to generate a 120-character string of letters and analyze the words that emerge and filter for meaning.

In the results, you can see the string generated, the words formed, the most interesting word found, the word family, its origin, and an associated writing prompt. This will be generated every morning at six a.m.

Our second tool takes on a harder task; we generate a longer string (not shown) and ask the API to find a longer, more complex word. This tool also returns the origin, word family, lineage and writing prompt. It generates a new word every morning at six a.m.

Due to monetary concerns, string and API use is limited and thus not every day will return a long word.

These fun little tools show that what we consider beauty, what we consider creative and free, is at its heart mathematical. What we create, is bound by rules inborn and unspoken. We are not making music, art, poetry, prose with an eye on the recursive, on similarity, on scaling, but they exist nonetheless.

Both tools are powered by Claude via the Anthropic API. The generated text is parsed in real time to find words in sequence. These are independent GoRhyme tools, not affiliated with or endorsed by Anthropic.