Lecture 7 Flashcards
Infants finding cues to word boundaries
stress patterns: in english most words start with a stressed syllable: if you assume that a stressed syllable indicates a word that should help you acquire words
stress-initial example
KINGdom
stress not initial
guiTAR
stressed syllable
TAR
Not all languages have the…. stress pattern
english-type
- with the word initial stress
How do you know what the properties of words are if you don’t know what the words are?
you can’t know that you have words with stress on the first syllable unless you know what some of the words are
statistical cues!
there are not actually spaces between words
even though in a familiar language there seems like there is
though silent boundaries may be an important cue in learning word boundaries
transition probabilities or conditional probabilities
just looking for patterns in a unfamiliar language: higher transition probability of one word transitioning into another word
what comes after “ba” alot of the time is “by” [baby]
implicitly keeping track of statistics used by infants to find out where word boundaries in their native language are
high transition probabilities
within words
most likely no word boundary
= 1
low transition probabilities
word boundaries exist
how do we find out if transition probabilities happen?
let’s do some math…
highest probability you can get is 1 (no word boundary) and the lowest is 0 (never happens)
it’s statistics
for “pa” if it’s just as likely to go to the syllables “li”, “go” and “nu” then the transition probability is
given “pa” what’s likely to come next?
1/3 or .33 transition probability for each syllable
Statistical Learning Experiement,
Saffrin, Aslin, and Newport, 1996
played made-up words to infants to assess how they were able to find word boundaries - all they had to go off of were the transition probabilities
1) played two different sounds: 3 syllable words with high within word transition probabilities and then with 3 syllable strings with lower transition probabilities (with word boundaries)
2) What do they listen to longer? - They listened to “PIRO.bi” (the low transition probability word) longer = indicates a novelty preference = infants reacting differently, suggesting they’re using statistical cues to tell the two types of words apart
What about stress?
stress vs. statistics
results from experiment Johnson and Jusczyk, 2001
results from Thiessen and Saffran (2003)
2001: tress suggested different word boundaries than statistics: found that “stress” wins
2003: Stats win at 7 months, Stress wins at 9 months
Segmentation precedes….
…stress!!!
Statistical learning
what’s getting learned? Experiment with Aslin, Saffran, Newport, 1996
in real language some words happen much more frequently
- made certain words occur 2x as often
- made word boundary crossing syllables more frequently than actual word (doti.go more than bidaku)
- they should listen longer to the “weird” thing (doti.go) than the one with the word boundary (bidaku) = YES they did
–> they are computing conditional probability rather than frequency
there are some syllable pairs that are not actually words
(thedog): it’s a problem because if all you’re keeping track of are syllables occuring together you might learn false words
conditional probably
given that syllable X has happened how likely is syllable Y to happen?
using the predictiveness of one syllable
works for language AS WELL AS for tone sequences, visual sequences = parsimonious and not language specific