Home▸Cognitive Science▸Statistical Speech Segmentation Lab

Statistical Speech Segmentation Lab

Watch a stream of nonsense syllables get split into 'words' purely from transitional-probability statistics, the same mechanism 8-month-old infants use (Saffran et al. 1996) to find word boundaries with no pauses, stress or grammar to help.

Cognitive Science3DModerate60 FPS

How does an infant find the boundaries between words in a continuous stream of speech, with no pauses to mark where one word ends and the next begins? Saffran, Aslin & Newport's classic 1996 experiment showed that 8-month-olds can do it from pure statistics: syllables inside a word always follow each other, so their transitional probability is high, while syllables that happen to sit across a word boundary co-occur far less reliably. This simulator builds a live artificial language out of four hidden three-syllable "words", streams them together with no acoustic cues, and runs an online statistical learner that tracks transitional probability syllable by syllable and drops a segmentation boundary whenever that probability dips below a threshold you control — with live readouts showing how close the learner's guesses land to the true, hidden word grid.

⚙ Under the hood

Streams a continuous artificial language of nonsense syllables and runs a live statistical learner that finds word boundaries purely from transitional probabilities, the mechanism behind infant word segmentation (Saffran et al. 1996).

psycholinguisticslanguage acquisitionstatistical learningcognitive scienceneuroscience

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install