This tool generates pseudo-text by replacing words from your input with new, non-semantic words that
mimic the original's phonetic and structural patterns. It does this using a character-level
Markov Chain model (Meyn & Tweedie, 2009) built from a source text you provide.
The Algorithm (N-grams)
Build Model: The generator reads your chosen source text (e.g., "The Little
Prince") and breaks it down into small overlapping chunks of characters called
n-grams (e.g., trigrams are 3 characters long). It learns which characters are
likely to follow any given chunk.
Generate Word: For each word in your input, it takes its length, picks a random
starting n-gram from the model, and builds a new word character-by-character by predicting the
most likely next character until the new word matches the original's length.
Validate: The new word is checked against the source dictionary. If it already
exists, the process repeats until a unique pseudo-word is found.
This tool is inspired by UniPseudo (Boris et al., 2024), a new tool to generate
readable pseudowords for psycholinguistic experiments. The present resource extends UniPseudo's
methodology by adding robust support for (1) longer text support in another web interface and (2)
Thai language support. We plan to extend this tool to Devanagari and Arabic, also using UniPseudo's
framework.
This tool was developed by Héctor Mangas (MSc), for Typotheque's 2025 Weight Survey. The study needed
a systematic approach to generate nonword stimuli in several scripts. The code for this page was
generated with the help of LLM Gemini 2.5 Pro.
How to Use
Paste Text: Add your content to the left panel.
Select Script: Choose "Latin", "Thai", "Tamil", "Urdu", "Hindi", or "Japanese" based on
your text. (The system will also try to auto-detect this!)
Import Source: In the right panel, import a source text (or paste your own) to
build the generation model. This is crucial for the generator to work.
Adjust & Copy: Use the settings to fine-tune the output, then copy or download
your new pseudo-text.