Adaptive Glyph Network
The same radial Numeraglyphs, now with memory. Blue links show fixed spelling-shape similarity; amber arrows learn which word tends to follow another. Explore whether repeated language can improve next-word predictions.
1. Teach the network
Repetition strengthens observed directional transitions. Try the alternative after learning the original: How this system learns from experience. Only confirmed words update memory.
2. Optional live dictation
This downloadable HTML opens locally for typed testing. Live microphone recognition may be blocked until hosted over HTTPS, permitted by the browser, and supported by the device. Recognition providers may process audio remotely.
3. Watch relationships change
4. Predict what comes next
Test without teaching
Reusing training text measures memorization, not generalization. Try a different sentence to test transfer.
Selected Numeraglyph
Learned outgoing links
Closest geometric glyphs
Shape scores use the original 82% radial-profile and 18% word-length metric. Sequence links use direct counts, not semantic embeddings.
This is an associative transition model, not a trained neural network. Prediction scores are observed fractions from a small dataset, not calibrated probabilities. No word meanings are inferred. No voice is recorded by the app; browser speech recognition may use external services. Default learning data exists only in page memory unless you opt into local storage or export it.