Module 3: Statistical learning and the word boundary problem

saxophone barbershop, hyphy bluegrass, drum and bass goa trance, dakar boom bap · 4:11

Listen on 93

Lyrics

[Verse 1]
Speech flows like a river with no breaks between the stones
No silence marks the borders where each meaning finds its home
Eight-month babies crack the code without a single pause to guide
They're counting up probabilities, statistics are their guide
When syllables stick together tight, the chances climb up high
But cross a word boundary and watch those numbers die

[Chorus]
Transitional probabilities, the dips reveal the seams
High inside, low outside, parsing language streams
Hip-po-cam-pus learning what comes next in every chain
Statistical segmentation, how the infant brain
Carves the flow to find the words hidden in the sound
Transitional probabilities, where word boundaries are found

[Verse 2]
Saffran played her nonsense streams for just two minutes straight
No acoustic cues available, pure statistics sealed their fate
The babies heard "tupiro" and "golabu" emerge
From continuous gibberish, their pattern-seeking urge
Distinguished real word units from the part-words at the edge
Predictive links decoded at eight months on the ledge

[Chorus]
Transitional probabilities, the dips reveal the seams
High inside, low outside, parsing language streams
Hip-po-cam-pus learning what comes next in every chain
Statistical segmentation, how the infant brain
Carves the flow to find the words hidden in the sound
Transitional probabilities, where word boundaries are found

[Bridge]
Not just for syllables alone, this machinery runs wide
Visual sequences and tones, motor patterns on all sides
Domain-general computation, not a language-only tool
The same stats that parse our speech make other patterns cool
MTL separates the items while CA fields connect
What tends to follow what—the chains that we expect

[Verse 3]
Fifteen years have changed the map of where this process lives
Cortical and frontal zones, but hippocampus gives
The rapid online learning power, temporal structure's key
Episodic memory systems set our word boundaries free
Striatal habits catch the patterns practiced day by day
While hippocampal prediction points the segmentation way

[Chorus]
Transitional probabilities, the dips reveal the seams
High inside, low outside, parsing language streams
Hip-po-cam-pus learning what comes next in every chain
Statistical segmentation, how the infant brain
Carves the flow to find the words hidden in the sound
Transitional probabilities, where word boundaries are found

[Outro]
From continuous streams to meaning, statistics light the path
Predictive associations, sequence-learning's math

← Module 2: Phonological learning and perceptual narrowing | Module 4: The hippocampus and fast mapping. The core of the course →