Fixation Sequences as Time Series: A Topological Approach to Dyslexia Detection

The real methodological bet here is treating fixation sequences as geometric objects rather than raw time series, which lets topological features capture reading rhythm irregularities that standard statistical summaries tend to flatten out. The Copenhagen Corpus is also worth flagging: it includes both native and non-native Danish readers, making cross-linguistic generalization a built-in test rather than an afterthought.

This sits at a distance from most recent Modelwire coverage, which has focused on LLM inference and evaluation. The closest conceptual neighbor is the DiscoTrace paper from April 16, which also compared human and model behavior through structured sequence representations, though DiscoTrace targets rhetorical strategy rather than clinical signals. More broadly, this work belongs to a quieter thread in the archive: using formal mathematical structure to extract meaning from behavioral traces that neural approaches tend to treat as noise.

The benchmark to track is whether this hybrid topological-statistical approach holds up when applied to corpora outside Scandinavian languages, particularly ones with orthographically irregular writing systems like English, where dyslexic fixation patterns are known to differ. If a replication on an English-language corpus like the Provo or MECO dataset shows comparable separation, the method has real generalization legs.