All the marvellous histories hidden in our languages.
Miraris - from Latin mirāri, second-person singular: "you wonder"
Etymology is fascinating
I'm often captivated by the hidden histories of the words we use every day. I like to understand how words were built, how they evolved into what they are today, and how and why their senses shift. I like to see how languages are connected to each other, how they influence each other, and what those threads may say about our common history and present.
I felt all that for a long time, but I had trouble finding tools, resources, or applications that let me simply see and grasp what I find so interesting about etymology. So I built Miraris. The goal of this app is to be a simple tool to use, that lets its users explore and see all the intriguing histories hidden in our languages.
All raw data comes from Wiktionary
All raw etymological data on Miraris is drawn from English Wiktionary, parsed via the structured dumps maintained at Kaikki.org.
This project could hardly exist without the work of the people contributing to Wiktionary. I thank them personally for their passion and generosity. I'm also grateful to everyone involved with Kaikki.org, whose work has been a great help.
Wiktionary's text is published under the Creative Commons Attribution-ShareAlike 4.0 license (CC BY-SA 4.0). Most of the data shown here has been cleaned, restructured, and summarized. In keeping with that license, the etymological content Miraris builds from Wiktionary data (the summaries, graphs, and translations) is itself made available under the same CC BY-SA 4.0 license.
- English Wiktionary: Primary source: etymologies, glosses, languages, cognates..
- Kaikki.org: Structured JSON dumps of Wiktionary, maintained by Tatu Ylonen.
Use of AI and notes on reliability
What the model does
Wiktionary's etymologies are written in prose. A language model (LLM) is used to turn prose into nodes and links to create etymology trees. It also reshapes the etymology texts into the “Word History” summary, and highlights the questions the sources leave open.
The system never relies on a model's own knowledge of etymology. A model is only given cleaned Wiktionary texts and asked to change their form. The tree it describes is then assembled and checked as a whole before it appears on the page.
An effort to keep it light
An effort is made in the design of Miraris to use small models as much as possible, while making sure the information stays accurate and relevant.
The same care goes into what is sent to a model and what is asked back. Texts are cleaned and prepared beforehand so that each request stays short, and a model is asked to return only what is needed. For a tree, that is a bare list of nodes and links, with everything else filled in from the database.
These efforts to use small models and to keep requests efficient are made to reduce the computing power and energy required to use Miraris.
The result for each word is also saved. Only the first search for a word calls the models, and every later visit reads the saved result.
Translations
Translations and near-equivalents work a little differently. For English words they come from Wiktionary whenever it has them. A model may fill in a few languages Wiktionary lacks, and those additions are marked with an asterisk. English Wiktionary only lists translations for English words, so for words in other languages they are suggested by a model, with a note under the list saying so, and rare words are usually left without any. This panel is the one place where Miraris leans on a model's own knowledge, of languages here, not of etymology.
Limits
The engine behind Miraris can still make mistakes, especially when the source text is ambiguous, incomplete, or for words with very deep etymologies and many connections. If you have doubts about anything on the app, you can inspect the raw etymological texts from Wiktionary by clicking any node on the graph, or cross-check with another source like Etymonline.
- Ambiguous source: When the Wiktionary entry hedges or contradicts itself, the model has to choose which reading to follow. Sometimes it picks the wrong one.
- Deep etymologies: Words with many ancestors, dialect labels, or contested branches are the most likely to come out simplified, or occasionally wrong.
The future of Miraris
I have plenty of ideas on how to improve this app, but I believe the most valuable insights will come from its users. You can share your feedback and suggestions on this feedback board, and vote for the ideas you want most. Don't hesitate to contact me, I will reply! My details are below.
Contact information
You can reach me by email or send me a direct message on X.
Email: max@miraris.app · X: @MaxouIsBuilding