WORD LEARNING

1. WORD LEARNING

2. A chicken-and-egg problem • You can't learn the language until you know the words BUT • You can't segment out the words in the speech input until you know what they are

3. How could a baby solve the dilemma? • Aslin, Newport & Saffran (1996, 1998) • Babies use the pattern of sounds within words to distinguish the ends of words • In Aslyn’s words, babies "pay attention to sounds that cohere within words, compared to the less predictive sounds that change as they span a word boundary.” • When that pattern breaks, the baby understands that a new word is about to start.

4. How could a baby solve the dilemma? • Part of a word [kn]… • … dle • … cer • … vass • … teen • … cel • … did [kn] • Whole word [kn] • … walk • … talk • … play • … cry • … hit • … eat • … think • ……………..

5. Transitional Probability (TP) What is the probability that X will be followed by _ ? X A TP(XA) = 1.0 A TP(XA) = 1/3 TP(XB) = 1/3 TP(XC) = 1/3 X B C

6. Saffran, Aslin & Newport 1996 • Fact: infants often listen longer to novel sounds rather than boring ones • Experiment: infants exposed 7- to 8-month-old infants to a nonsense language for two minute • Question: will infant learn the regularities of the nonsense language?

7. Saffran, Aslin & Newport 1996 • Nonsense language based on 12 different syllables • Has four tri-syllabic words: word = s1-s2-s3 • pabiku • tibudo • golatu • daropi • Presented as a string of nonsense syllables with no pauses indicating word endings pabikutibudogolatudaropitibudodaropipabiku…

8. Saffran, Aslin & Newport 1996 • Transitional Probabilities (TP) of the nonsense language • TP of between within-word syllables (i.e. s1-s2 or s2-s3): 1.0 • TP of between between-word syllables: 1/3 (s3-s1, each initial syllable of a word can follow other 3 words of the language, i.e. other 3 syllables)

9. Saffran, Aslin & Newport 1996 • Testing phase • 4 items in total • 2 of the 'words' from familiarisation, e.g. pabiku & tibudo • 2 ‘partwords’, e.g. tudaro & pigola

10. Saffran, Aslin & Newport 1996 • Results (n=30) Mean Listening times (seconds) Words part-words Matched-pairs t test 6.78 7.36 t(29) = 2.1, P < 0.05

11. Saffran, Aslin & Newport 1996 • Interpretation Infants can distinguish between words and part-words after two minutes of exposure • Infants are sensitive distributional properties of the language

12. Learning names for objects Stager & Werker, 1997

13. A Straightforward Learning Scenario: • By 1 year of age children know all phonemes in their language • They realize that different objects must have different names (where the difference can be as small as one phoneme) • If so we expect them to have no problem learning word names, including minimal pairs of words (e.g. bear – pear, big – pig)

14. Stager & Werker 1997 • Novel objects • Novel names: ‘bih’, ‘dih’, ‘lif’, ‘neem’

15. Stager & Werker 1997: Switch task • Word pairs: ‘bih’ vs. ‘dih’ ‘lif’ vs. ‘neem’

16. Word learning results • Exp 2 vs 4

17. Paradox! • 14-month olds fail on minimal pairs of words • But they do know minimal pairs of sounds (as shown in the task that does not require word-learning) • They do know the sounds but they fail to use the detail needed for minimal pairs to store words in memory

18. Resource Limitation Hypothesis • Stager & Werker (1997) • The complex nature of word learning limits use of the available phonetic information. For a novice word learner, forging a link between a label and an object is a computationally demanding task. • Thus, the resources available for attending to the fine phonetic detail of the word are limited.