Infant perception phoneme and words

Question 1

How early can infants extract words from speech stream?

Answer 1

Juscyk and Aslin 1995: familiarisation phase: passage that repeats words within context of speech
After 2 minute presentation, 7.5month infants able to recognise recurring patterns in stream.
Appeared to show preferential looking with key words when played in isolation with novel
did not look longer at ‘incorrect’ versions of the words: accurate acoustic phonetic representations.

Question 2

What mechanisms are used for word segmentation?

Answer 2

words as anchors, stress patterns, phonotactic constraints, function and content words, statistical learning (SL) for regularities.

Question 3

frequently heard isolated words as anchors

Answer 3

only 9% of mothers words are isolated at 9-15m (brent and siskind 2001)
75% of 18mo first words had previously appeared in isolation in mothers speech- facilitate SL within longer utterances (lew-williams 2011)

Question 4

sensitivity to:

dominant stress patterns of the language

Answer 4

english= strong weak pattern for 90% words (Cutler and carter 1987); use to predict start of word.
Jusczyk 1999- Nativity to stress patterns develops at 6-9 months, recognise regular pattern words at 7.5m, By 10m they recognise familiarised weak/strong words in fluent so stress strategy becomes less important
— doesn’t explain how this is learnt; poverty of stimulus not consistent enough to learn?!

Question 5

Sensitivity to phonotactic constraints

Answer 5

these develop @ 6-9m, frequent phono structures preferred by infants aged 9 months over infrequent structures (jusczyk 94)
Accumulate frequencies of occurrence; but some are very ambiguous and coarticulation

Question 6

functional and content word discrimination

Answer 6

due to acoustic or phonoloigcal patterns rather than just exposure
shi and werker 2011: prefer function rather than content words in an unfamiliar language
Shi 2006: prefer novel funciton word rather than mispronounced function word
- sensitive to stress; aids segmentation (shi and werker 2001)

Question 7

Saffran 2003

Answer 7

SL: use properties to discover structure, including sound patterns, words and beginnings of grammar

Question 8

Saffran 1996 paradimg

Answer 8

1) artificial lagnuage with no segmentation cues; 2 min presentation, HTP
2) played segments which did or did not occur
3) track frequency of trisyllabic words within 2 minute stream
4) track probabilities of co-occurence of pairs of syllables
- very controlled

Question 9

syllable co-occurrence best explains transitional probability
- test design?

Answer 9

Stronger transition with pretty/baby than pre/ttyba/by.
Test design (Aslin 1998):  frequency is manipulated so frequency of occurrence varies  with greater transitional probability for word strings over part words: infant still showed novelty effect for part words

Question 10

Is this specific to language?

Answer 10

NO: comparison of SL for speech and tone, evidence of adult and infant learning tone sequences words are greater than part words:- SL is general to learning not specific

Question 11

more ecologically valid technique than artificial language?

Answer 11

pelucchi 2009: stat learning from naturally occuring stimulli from a novel language (italian)- similar results ot artificial; stimuli with intonation patterns resembling those of IDS show better learning
- learning is reinforced by prosodic patterns of exposure

Question 12

Infants DO/DON’T show evidence of early rule learning

Answer 12

marcus 99 // johnson and tyler 2010

Question 13

Marcus 99

Answer 13

grammatical level of learning, 7mo familiarsation phase with ABA grammar, tested with patterns that violated the patterns
Infants showed novelty effect for the irregular sentences- learnt rule and re applied (not evo-valid)

Question 14

Johnson and Tyler 2010

Answer 14

eco valid: 5.5-8mo: artificial lang containing four words of uniform length OR four words varying in length.
Neither age group succeeded with language containing words of varying length despite word boundaries equally strong
- so SL not as robust?

Question 15

Chomsky, gambell and yang 2005

Answer 15

SL does not scale up to realistic settings of langauge acq; transitional probabilities do not match with IDS which contains mainly monosyllabic words
SL strengthens the case for universal grammar: must know the relevant unit of info to track
- learning only possible with prior assumptions – poverty of stimulus

Question 16

Saffran 1999

Answer 16

countered chomsky: common SL for speech and tone segmentation are acquired by same mechs for segmentation- domain- general mechanisms.

Question 17

when is meaning learnt?

Answer 17

By 6 m: infant associate small number of high freq words with referent.
By 1y: can associate frequent words
14-20m sensitive to mispronunciations and thus phonetic detail (swingley and aslin 00/02)

Question 18

Switch paradigm (Werker 1988)

Answer 18

Present parings (novel object and name) 7-10 times> habituation
- test phase: association = longer looking when switched

Question 19

Werker 1988 findings?

Answer 19

at 14m infants learn pairing of 2 novel words with 2 different objects but fail with minimal contrasts
At 8-14m can discriminate
- cognitive demands?
reduced phonetic sensitivity to reduce load so more managable

Question 20

Wojcik and saffran 2013:

Answer 20

27mo taught to associate each nonsense word with image and tested with audio only- longer looking at pairs denoting visually similar objects

• toddlers learn new words by encoding relationships among referents as well as associations.
• technique clearly shows networks

Question 21

Zamuner 2009

Answer 21

phonotactic probabilities in dutch controlling for word position- Correlations were found between vocabulary size and segment repetition accuracy
=Phonological representations are mediated not only by children’s developing vocabs but also the structure of childrens emerging lexicon.

Question 22

Kohlaguina and shi 2012

Answer 22

• 11m and 14m; Sentences in russian
• In training word of each sentence was transformed following a consistnet pattern (ABC-BAC)
• Test phase: learnt novel sentences respected the trained rule and others that violated (ABC-ACB)
  =14mo but not 11mo showed evidence of abstract rule generalisation to novel instances

Question 23

not just for language.. can be tones for boundaries OR…

Answer 23

Stat defined visual patterns (fiser and aslin 2002, kirkham et al 2002)