1. Announcements Switching lecture schedule:
Move Pragmatics Unit to AFTER Bilingualism Unit to accommodate guest speaker on Nov. 27th (Tuesday)
If we have time after lecture today, we’ll discuss some posted questions.

2. Psy1302 Psychology of Language
Language Acquisition I
Lecture 17

3. Language Acquisition
[The acquisition of language] is doubtless the greatest intellectual feat any one of us is ever required to perform.
(L. Bloomfield, Language, 1933, p. 29)

4. Reading Assignment Fisher & Gleitman (2002)
I. Outline of the task of language learning
II. Where language learning begins
Categorization of Speech Sounds
Segmentation of Spoken Word
Role of Sound in Syntactic Analysis
Distributional Analysis and Discovery of Syntax
II. Meanings
Primitive Categories of Experience
Compositional Meaning
Interactions between linguistic and conceptual categories
IV. Forms to meaning
Mapping problem
Concrete words first
Old words make new words easier to learn
V. Where learning ends

5. Eimas et al.’s Study High Amplitude Sucking Procedure
Review: Where we left off for lectures on Speech Perception
Eimas et al.’s Study
High Amplitude Sucking Procedure
What did this paradigm tell us about infants’ ability to discriminate speech sounds in the world’s languages?

6. Werker et al.’s Studies Headturn Procedure
Review: Where we left off for lectures on Speech Perception
Werker et al.’s Studies
Headturn Procedure
What did this paradigm tell us about when children lose discrimination ability of non-native phonemes

7. A New Puzzle: Learning Native Phones
How do children learn which phonetic sounds are the phonemes in their language?
For example (minimal pairs):
‘bear” and “pear” or
“rent” and “lent” are two different words?
What information could children use to learn the phonemes of their native language???

8. One Possibility: Semantic Learning
A New Puzzle: Learning Native Phones
One Possibility: Semantic Learning
Learning word meanings drives phonological reorganization
Children notice changing some features changes meaning in some cases
/tip/ vs. /dip/
They also notice changing some features does not change meaning in some cases
/tip/ vs. /tip/ (Hindi contrast) ,

9. One Possibility: Semantic Learning
A New Puzzle: Learning Native Phones
One Possibility: Semantic Learning
Learning word meanings drives phonological reorganization
For reorganization to work, requires:
Noticing minimal pairs and meaning differences
Learning many words
Problem: 10 months-old children don’t know that many words!

10. Review
Voice Onset Time
Production of words with /d/ or /t/ have different VOT

11. Dutch
Easter
Armenian
Spanish
Thai
Hungarian
Korean
Tamil
Hindi
Cantonese
Marathi
English

12. Another Possibility: Distributional Learning
A New Puzzle: Learning Native Phones
Another Possibility: Distributional Learning
Distributional characteristics of input (i.e. speech of native speakers) provide cues to the phonemic categories
E.g. 2 clusters

13. Another Possibility: Distributional Learning
A New Puzzle: Learning Native Phones
Another Possibility: Distributional Learning
1 cluster vs. 2 clusters d t

14. Maye, Werker, & Gerken (2002) (tested 6 & 8 months-old)
A New Puzzle: Learning Native Phones
Maye, Werker, & Gerken (2002) (tested 6 & 8 months-old)
Created synthetic sounds ranging from [da] to [ta].

15. Maye, Werker, & Gerken (2002) (tested 6 & 8 months-old)
A New Puzzle: Learning Native Phones
Maye, Werker, & Gerken (2002) (tested 6 & 8 months-old)
Familiarized infants to one of two sets
Bimodal Set: Sounds on the ends near [da] and [ta].
Unimodal Set: Sounds in the middle.
Test preference for:
… (Alternating) vs … (Non-alternating) Stimuli

16. Maye, Werker, & Gerken (2002) (tested 6 & 8 months-old)
A New Puzzle: Learning Native Phones
Maye, Werker, & Gerken (2002) (tested 6 & 8 months-old) … = =
<
<
Infants trained on the Bimodal had a novelty preference for non-alternating trials.
Infants trained on the Unimodal did not prefer/dis-prefer one over the other.

17. Summary Infant’s initial lexicon is non-existent to small.
A New Puzzle: Learning Native Phones
Summary
Infant’s initial lexicon is non-existent to small.
Makes it difficult to attribute lexical knowledge as source of initial phonological reorganization
Perhaps distributional information may be a first step into initial phonological reorganization
Learning the phonemes of one’s language could then help learn words & word meanings…
Big Q: how do you know which cues to analyze? E.g. VOT as opposed to some other cues?

18. Posting @ Class Discussion Forum (10/12)
A New Puzzle: Learning Native Phones
Digression
Class Discussion Forum (10/12)
Minimal Pairs and Word Learning:
Notice anything about the ages tested in these papers and when the Werker paper says that sensitivity to non-native phones disappears? Want to comment?
Monolingual:
Bilingual:

19. The “Switch” Task Habituation Phase Test Phase Same Switch
A New Puzzle: Learning Native Phones
Digression
The “Switch” Task
Habituation Phase Test Phase
Same Switch
“lif” “neem” “lif” “neem”
Werker, Cohen, Lloyd, Casasola, & Stager, Dev Psych, 1998

20. The “Switch” Task Results
A New Puzzle: Learning Native Phones
Digression
The “Switch” Task Results
Werker, Cohen, Lloyd, Casasola, & Stager, Dev. Psych. 1998

21. Switch task with Minimal Pairs
A New Puzzle: Learning Native Phones
Digression
Switch task with Minimal Pairs
Habituation Phase Test Phase
Same Switch
“bih” “dih” “bih” “dih”
Stager & Werker, Nature, 1997

22. Switch Task, Minimal Pairs Result
A New Puzzle: Learning Native Phones
Digression
Switch Task, Minimal Pairs Result
Stager & Werker, Nature, 1997, Werker, et al, Infancy, 2002; Pater, et al, Language, 2004

23. Posting @ Class Discussion Forum (10/12)
A New Puzzle: Learning Native Phones
Digression
Class Discussion Forum (10/12)
Minimal Pairs and Word Learning:
Notice anything about the ages tested in these papers and when the Werker paper says that sensitivity to non-native phones disappears? Want to comment?
Monolingual:
Bilingual:

24. Word segmentation problem
Old Puzzle, But Yet Another Issue: Word Segmentation
Word segmentation problem
How do we as adults know where the word boundaries are located?
We make use of lexical knowledge.

25. Word segmentation problem
Old Puzzle, But Yet Another Issue: Word Segmentation
Word segmentation problem
What about infants who have none or few words in their lexicon?

26. Word Segmentation Problem
Old Puzzle, But Yet Another Issue: Word Segmentation
Word Segmentation Problem
Infants need to extract words from fluent speech in order to build a lexicon.
How???
Cues recruited for segmentations:
Rhythmic cues
Metrical Stress
Phonotactic cues
Context sensitive allophonic cues
Statistical distributional properties
Transitional Probabilities of Syllables
Allophonic cues:
/vt/ rare within word, more common across words
/d/ in red ice and dice different

27. Rhythmic Cues Stress Pattern in English:
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues – stress pattern
Rhythmic Cues
Stress Pattern in English:
Cutler & Norris (1988): 95% of disyllabic English words (actually heard) are Strong-Weak.
Common (Strong-Weak Stress): Button, Table
Rare (Weak-Strong Stress): Guitar, Surprise
Do (English) infants take strong-weak as cue for segmentation?

29. Experimental setup

30. Rhythmic Cues Jusczyk, Houston & Newsome (1999)
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues – stress pattern
Rhythmic Cues Jusczyk, Houston & Newsome (1999)
7.5 & 10 month-old infants
familiarized to Word, and test preference for passage w/ over passage w/o target word
Familiarized to Word in Passage, and test preference for target word over another (novel) word

31. Rhythmic Cues Jusczyk, Houston & Newsome (1999)
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues – stress pattern
Rhythmic Cues Jusczyk, Houston & Newsome (1999)
7.5 & 10 month-old infants familiarized to
Strong-Weak Word:
Kingdom… Kingdom… Kingdom…
Test on passages. Compare preference for passages w/ or w/o target word.
Kingdom passage
Your kingdom is in a faraway place. The prince used to sail to that kingdom when he came home from school. One day he saw a ghost in this old kingdom…
Hamlet passage
Your hamlet lies just over the hill. Far away from here near the sea is an old hamlet. People from the hamlet like to fish. Another hamlet is in the country…

32. Rhythmic Cues Jusczyk, Houston & Newsome (1999)
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues – stress pattern
Rhythmic Cues Jusczyk, Houston & Newsome (1999)
7.5 & 10 month-old infants familiarized to
Strong-Weak Word in passage:
Your kingdom is in a …
Test on words. Compare preference of target word and another word
Kingdom passage
Your kingdom is in a faraway place. The prince used to sail to that kingdom when he came home from school. One day he saw a ghost in this old kingdom…
Hamlet passage
Your hamlet lies just over the hill. Far away from here near the sea is an old hamlet. People from the hamlet like to fish. Another hamlet is in the country…

33. 7.5 months hear kingdom  parse as kingdom and not king
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues – stress pattern
7.5 months
S/W (strong-weak): kingdom
king
king
king
king
hear kingdom  parse as kingdom and not king

34. Rhythmic Cues Jusczyk, Houston & Newsome (1999)
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues – stress pattern
Rhythmic Cues Jusczyk, Houston & Newsome (1999)
7.5 & 10 month-old infants familiarized to
Weak-Strong:
Guitar… Guitar… Guitar…
Test on passages. Compare preference for passages w/ or w/o target word.
Guitar passage
The man put away his old guitar. Your guitar is in the studio. That red guitar is brand new. The pink guitar is mine….
Device passage
Your device can do a lot. Her device only fixes things. My new red device makes ice cream. The pink device sews clothes….

35. 7.5 months hear “guitar”  parse as tar and not guitar
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues – stress pattern
7.5 months
W/S (weak-strong): guitar
tar
tar
guitar is
tar
guitar is
taris
hear “guitar”  parse as tar and not guitar
hear “guitar is”  parse as taris and not guitar or tar

36. 10 months Older children not as dependent on strong-weak strategy.
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues – stress pattern
10 months
W/S (weak-strong): guitar
tar
guitar is
taris
guitar is
Older children not as dependent on strong-weak strategy.
hear “guitar”  parse as guitar

37. Rhythmic Cues Jusczyk, Houston & Newsome (1999)
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues Jusczyk, Houston & Newsome (1999)
Rhythmic Cues – stress pattern
For SW: Both 7.5 & 10 month-olds gave the correct parse
For WS: Only the 10 month-olds gave the correct parse
Young (English-learning) infants rely on SW pattern for segmenting speech

38. Phonotactics Mattys & Jusczyk (2001)
Old Puzzle, But Yet Another Issue: Word Segmentation
Phonotactic Cues
Phonotactics Mattys & Jusczyk (2001)
Headturn Preference Procedure again
9 mos. familiarized to passage with 1 of 2
novel words
Gaffe
Tove
The novel words were either flanked by
good between-word cues or
bad between-word cues.
(e.g. /vt/ is rare within a word, but more common across words)
Test word in isolation (e.g. gaffe, gaffe, gaffe…)

39. Phonotactics Mattys & Jusczyk (2001)
Old Puzzle, But Yet Another Issue: Word Segmentation
Rhythmic Cues – stress pattern
Phonotactics Mattys & Jusczyk (2001)
gaffe
tove
Words flanked by good between-word cues or bad between-word cues.
C CVC C
Novel word
Offset of
Another word
Onset of
Another word

40. Statistical Learning Saffran, Aslin, & Newport (1996)
Old Puzzle, But Yet Another Issue: Word Segmentation
Syllable Transitional Probabilities
Statistical Learning Saffran, Aslin, & Newport (1996)
Transitional Probabilities of Syllables
Likelihood of a syllable (B) following another (A)
Probability of B given A
freq AB
freq A
pr B|A =
Online demonstration of statistical analysis
Debate on whether this aspect of language is species-specific – relevant to issues of language as a domain-specific or domain general phenomenon. A B

41. PRE TTY BA BY Statistical Learning
Old Puzzle, But Yet Another Issue: Word Segmentation
Syllable Transitional Probabilities
Statistical Learning
Syllable Transitional Probabilities
High likelihood
High likelihood
PRE TTY BA BY
Low likelihood
Continuations within words are systematic
Continuations between words are arbitrary

42. Transitional probabilities
Old Puzzle, But Yet Another Issue: Word Segmentation
Syllable Transitional Probabilities
Transitional probabilities
PRETTY BABY
(freq) pretty
(freq) pre
.80
versus
(freq) tyba
(freq) ty
.0002

43. Statistical Learning Demo

44. tokibugikobagopilatipolutokibu
gopilatipolutokibugikobagopila
gikobatokibugopilatipolugikoba
tipolugikobatipolugopilatipolu
tokibugopilatipolutokibugopila
tipolutokibugopilagikobatipolu
tokibugopilagikobatipolugikoba
tipolugikobatipolutokibugikoba
gopilatipolugikobatokibugopila

45. tokibugikobagopilatipolutokibu
gopilatipolutokibugikobagopila
gikobatokibugopilatipolugikoba
tipolugikobatipolugopilatipolu
tokibugopilatipolutokibugopila
tipolutokibugopilagikobatipolu
tokibugopilagikobatipolugikoba
tipolugikobatipolutokibugikoba
gopilatipolugikobatokibugopila

46. Statistical Learning Saffran, Aslin, & Newport (1996)
Old Puzzle, But Yet Another Issue: Word Segmentation
Syllable Transitional Probabilities
Statistical Learning Saffran, Aslin, & Newport (1996)
Headturn preference paradigm
Familiarization:
2 minute exposure to an artificial language
4 3 syllable words: tokibu, gopila, gikoba, tipolu
Synthesized speech.
Only statistical cues to word boundaries
No prosodic information, coarticulation cues
Tested isolated word and part-word (sequences spanning word boundaries)

47. tokibugikobagopilatipolutokibu
gopilatipolutokibugikobagopila
gikobatokibugopilatipolugikoba
tipolugikobatipolugopilatipolu
tokibugopilatipolutokibugopila
tipolutokibugopilagikobatipolu
tokibugopilagikobatipolugikoba
tipolugikobatipolutokibugikoba
gopilatipolugikobatokibugopila

48. tokibugikobagopilatipolutokibu
gopilatipolutokibugikobagopila
gikobatokibugopilatipolugikoba
tipolugikobatipolugopilatipolu
tokibugopilatipolutokibugopila
tipolutokibugopilagikobatipolu
tokibugopilagikobatipolugikoba
tipolugikobatipolutokibugikoba
gopilatipolugikobatokibugopila

49. TO KI BU GI KO BA Test: TOKIBU vs. BUGIKO Word Part Word
Old Puzzle, But Yet Another Issue: Word Segmentation
Syllable Transitional Probabilities
Test: TOKIBU vs. BUGIKO
Word
Part Word TO KI BU GI KO BA

50. Results Words Part-words * Looking times (sec) 8 6 4 2
Old Puzzle, But Yet Another Issue: Word Segmentation
Syllable Transitional Probabilities
Results
Looking times (sec) 2 4 6 8
Words
Part-words *
Infants can use statistical cues to find word boundaries

51. Word Segmentation Problem
Old Puzzle, But Yet Another Issue: Word Segmentation
Word Segmentation Problem
Infants need to extract words from fluent speech in order to build a lexicon.
Considerable evidence that this ability develops in infants between months of age.
Cues recruited for segmentations:
Rhythmic cues
Metrical Stress
Phonotactic cues
Context sensitive allophonic cues
Statistical distributional properties
Transitional Probabilities of Syllables
Allophonic cues:
/vt/ rare within word, more common across words
/d/ in red ice and dice different

52. Old Puzzle, But Yet Another Issue: Word Segmentation
Infants detect and make use of a variety of probabilistic cues to determine word boundaries
They do so to construct a lexicon, and make use of the cues even before attaching meanings to the words
Which cues are more potent?
Reliance on one cue over another will depend on:
The language being learned
Sophistication of the learner

53. Is there long-term retention of these extracted sound patterns
Old Puzzle, But Yet Another Issue: Word Segmentation
Is there long-term retention of these extracted sound patterns
Houston, Jusczyk and Tager (1998)
Familiarized 7.5 mo. old infants with a pair of words and tested them on passages with and without familiarized words after 24 hours.
Results: No difference b/w the infants tested with delay and without delay.

54. Old Puzzle, But Yet Another Issue: Word Segmentation
Continued..
Jusczyk and Hohne (1997):8 month old infants heard audio versions of 3 stories for 2 weeks.
After 2 weeks, infants were tested on words and foils (matching phonetic characteristics and freq. of occurrence of words).
Infants who had heard the stories listened longer to the story words than the foils.
Control group: No preference

55. Old Puzzle, But Yet Another Issue: Word Segmentation
Conclusion
Infants store information about sound patterns of lexical items even before they attach meaning to these words.

56. Reading Assignment Fisher & Gleitman (2002)
I. Outline of the task of language learning
II. Where language learning begins
Categorization of Speech Sounds
Segmentation of Spoken Word
Role of Sound in Syntactic Analysis
Distributional Analysis and Discovery of Syntax
II. Meanings
Primitive Categories of Experience
Compositional Meaning
Interactions between linguistic and conceptual categories
IV. Forms to meaning
Mapping problem
Concrete words first
Old words make new words easier to learn
V. Where learning ends

57. Richie’s Posting @ Class Discussion Forum
What types of explanations have been offered for why languages sound so different?
Similarly, what do psycholinguists predict about the future of languages? Do they see a convergence towards one language as technology makes geographical borders increasingly non-existent? What does the future of language hold for us?

58. Courtney’s Posting @ Class Discussion Forum
If, when we speak, we produce sentences through the three steps we discussed in lecture on Tuesday, how are those steps manipulated, or how are they upheld, when we "speak" to ourselves? How do we go from the semantic to the articulatory stage when we are reading aloud? I know I've made incorrect "articulations," or at least word assumptions when reading. Does this mean that when we do make mistakes in speaking (through exchanges, anticipation, etc.) that the error occurs not in the physical articulation but in the mental processing stage?