Language Flashcards
What are the building blocks of language? (state)
Words
Phonemes (sound units)
Morphemes (smallest units of language)
Syllables (rhythmic units)
Stress (relative emphasis of syllables)
Words (language building blocks)
Representations stored in mental lexicon:
- catalogue of words - mental dictionary
- average person has 40,000-50,000 words
- ML contains spelling, pronunciation, meaning, grammar category
Phonemes (language building blocks)
- sound units of language - smallest units of speech that allow word discrimination
- ~100 phonemes across world - ~40 in English
- words not monolithic units of info - there are subcomponents
- Spoonerisms (William Spooner)
- in toast the the queen: ‘queer old dean’ not ‘dear old queen’
- shows you can replace sound units
- phonemes are cognitively real
- Kuhl et al., (1992) - infants can distinguish between most phonemes
- tune in to native language by age 1
Morphemes (language building blocks)
- Smallest units that carry meaning
- ’s’, ‘ing’, ‘ly’, ‘ness’, ‘er’
- words can be morphologically complex or simple
- complex = has multiple morphemes
- morphological overlap (share morpheme)
- will be faster to recognise the second
- stored in ML somewhere?
- morphological stranding = speech error where morphemes stay in the same place even if words swap
- ‘a geek for books’ –> ‘a book for geeks’
Syllables (language building blocks)
- rhythmic units - one vowel with or without surrounding consonants
- expletive infixiation rule
- can only insert expletive if there is a certain syllabic structure
- ‘fan-fucking-tastic’
- McCarthy (1982) - multiple syllables, word has main stress preceded by secondary stress
Stress (language building blocks)
- relative emphasis of certain syllables - can alter meaning (CON-tent vs con-TENT)
- Cappa et al., (1997) - aphasic CV
- can produce individual phonemes but the stress is wrong
- produces speech errors when stress fell on wrong syllables
- dissociates phonemes from stress
Cognitive neuroscience of language:
- Broca’s area
- Wernicke’s area
Broca (1824-1880)
- impaired language production, relatively intact comprehension (patient Tan)
- left inferior frontal gyrus
Wernicke (1848-1905)
- fluent (but disordered) production, impaired comprehenaion
- left posterial temporal lobe
Classical language model
- Wernicke-Geschwind model
Proposed by Wernicke (1874), extended by Geschwind (1970)
spoken word –> area 41 (auditory cortex) –> Wernicke’s area –> hear + comprehend word
cognition –> Wernicke’s area –> Broca’s area –> facial area of motor cortex –> cranial nerves –> speak
written word –> area 17 (V1) –> areas 18+19 (V2) –> area 39 (angular gyrus) –> Wernicke’s area –> read
(NB: typically left-lateralised)
Current views of language model
(neurobiological architecture)
Commonly activated regions:
- left inferior frontal gyrus (Broca’s area)
- superior, medial, inferior temporal gyri (L+R)
- frontal regions in L hemisphere
White-matter tracts:
- arcuate fasiculus –> links temporal + frontal regions (dorsal stream)
- sound –> articulatory representations
- L more prominent
- extreme capsule –> links frontal to temporal regions (ventral stream)
- speech –> conceptual representations
- Hickok & Poeppel, 2004
- some degree of bidirectionality in D+V streams
Vigneau et al., (2006) - meta-analysis of 128 imaging studies
- prominent L frontotemporal network (also significant R hemisphere involvement)
- highly interconnected regions - LH dominant but RH too
- different parts of network activated based on task + input modality
- phonological clusters, semantic clusters, sentence clusters
- all partially overlapping
- no strict modularity - interactions are key
Properties of written language
- writing systems
- role of regularity
Writing systems:
- logographic = unique symbol per word/morpheme (Chinese)
- syllabic = symbol for syllable (Japanese)
- alphabetic = unique unit for each phoneme (english)
- Dehaene (2009) –> diverse but all share visual features
- recurring shapes, contrasting contours, average of 3 strokes per character
Role of regularity
- Deep orthography (English/Hebrew)
- letters/groups of letters have different sounds in different contexts
- Shallow orthography (Finnish/Spanish)
- consistent correspondence between letters and phonemes
Visual word recognition process:
- Process
- state components
Process:
- extract from visual input
- letter recognition
- orthographic lexicon AND/OR grapheme –> phoneme conversion
Components:
- Eye movements
- Letter recognition
- Orthographic lexicon
- Grapheme –> phoneme conversion
Eye movements (visual word recognition)
- fixation and saccades
- fixation brings text into foveal vision
- high concentration of photoreceptors
- average fixation = 200-250ms
- average saccase length = 8 letter hop
- 10-15% of time they move backwards
- fixation brings text into foveal vision
Letter recognition (visual word recognition)
- recognise visual characteristics THEN identify
- Miozzo & Carmazza (1998) - alexic patient RV
- fine with visual characteristics
- impaired at identifying words
- alexic patients with lesions in L posterior regions - issue with identity
- Visual word form area (L fusiform gyrus)
- extracts identity of letter string
- regardless of size/shape/position/language
- Dehaene et al., (2001): visual priming
- if prime with target word - reaction time decreases
- subconscious early processing (implicit memory)
Orthographic lexicon (visual word recognition)
stores representation of spelling
- activated when we read a familiar word
- then obtain meaning from semantic systems
Grapheme to phoneme conversion (visual word recognition)
When reading novel words or pseudowords –> need to assemble the pronunciation from its letters
- suggests word-processing models need partially distinctive mechanisms for regular vs irregular vs pseudowords
Dual-route model of written language processing
- Coltheart et al., (2001)
- evidence for separate systems
- problems
Lexical route
- faster for words
- frequency-weighted
- faster for regular words
Non-lexical route
- faster for pseudowords and irregular words
- assemble from letters and rules
- if issue: problem with assembly
- so retrieve closes real world word instead
Evidence for separate:
- Patterson (1982) - AM –> problems with pseudowords
- lexicalisation –> read it like closest legical item
- 83-95% accuracy on normal words
- 0-37% accuracy on pseudowords
- Shallice et al., (1983) - HTR –> problems with irregular words
- surface dyslexia
- 79% accuracy on regular, 84% on pseudo, 48% on irregular
PROBLEM:
- DRC fully hardwired –> can’t learn new rules (for non-lexical)
- focus on English
- doesn’t explain HOW implemented in brain
Triangle models (written word processing)
- overview
- pseudowords
- irregular words
- experimental data
Seidenberg & McClelland (1989); Harm & Seidenberg (2004)
- orthography (letters); meaning (semantic units); phonology (sounds)
- visual input –> O –> M/P
- P –> output
- same mechanism for words + pseudowords; irregular words + regular words
- system picks up correspondences between specific orthographic units and phonological units
- strong connections between units more often co-activated
Pseudowords –> O + P (no M because not real) –> if P damaged, need to rely on O-M correspondence –> so lexicalisation
Irregular words –> O + M (no P because irregular) –> if O damaged, need to reply on P-M (phonology) - so regularisation
BUT: focus on English AND doesn’t explain HOW implemented in brain
Written word processing in the brain
Marinkovic et al., (2003): MEG
- activation starts in occipital lobe, spreads temporally
- occipital-temporal junction
- spreads temporally down ventral route (meaning) + frontal (top-down processing)
Dehaene et al., (2009):
- ventral stream:
- frontotemporal network (meaning)
- bidirectional (top-down, bottom-up)
- frontotemporal network (meaning)
- dorsal stream:
- occipital –> parietal –> articulation/pronunciation
- occipital lobe activation - low-level processing visual input
- VWFA - binds lower-level processing to language network
Neuronal recycling hypothesis
Dehaene et al., (2009)
- really high correlation between naturally-occuring image fragments + their frequency of occurence in written symbols
- maybe we use visual image processing for language
- reading co-opted evolutionarily-older brain functions - for visual object processing
- explains why symbols are what they are
- explains why we’re so good at reading even though evolutionarily new
Properties of spoken language
- problems to solve
- speech recognition - extracting invariant representations from continuous variable input
- speech signal = continuous, distributed in time, fast-fading and variable (IDs)
- word-segmentation problem - where do words start + end
- word co-articulation problem - blending of sounds from one word to the next
- child’s paradox - how do children separate words if they don’t know them
Major elements of speech recognition (state them)
- word segmentation
- lexical selection
- access to meaning
- context effects
Word segmentation (speech recognition)
- MSS
- evidence
- evaluation
- Cutler & Norris (1988) - metrical segmentation strategy
- stressed syllables at onset of words
- continuous speech segmented at stress syllables
- stressed syllables more likely to be content words (have full vowels)
- unstressed syllables more likely to be grammatical words
- Cutler & Carter (1987) - evidence supporting MSS:
- English syllable distribution
- 3/4 of time: stressed = content, unstresses = grammatical
Evaluation:
- not infallible - just a strategy
- need other sources of info to supplement
- language-specific
- solves child’s paradox
Lexical selection (speech recognition)
- shadowing paradigm
- gating paradigm
- cohorts
- salience of onset
- late activation
- Searching process - determine best fit between input + lexical representations
- fast - words in context can be recognised 175-200ms from onset (only part of content presented)
- Marslen-Wilson (1975) - shadowing paradigm
- hear sentence + repeat
- sentence has some mispronunciations
- people correct when repeating even before full presentation
- not just representation
- Tyler & Wessels (1983) - gating paradigm
- presented with parts of a word - get progressively longer
- decreasing cohorts of possible endings as word gets longer
- uniqueness point = point at which the word can be disambiguated from all others in the cohort
- Marslen-Wilson & Welsh (1978) - cohort models
- when we hear a word we start a parallel search consistent with the onset
- as more info comes in, words drop off + cohort gets narrower
- Tyler (1984) - set size decrease happens faster in context
- Marslen-Wilson (1980) - ~150ms needed to set up a cohort
- Cole & Jakimik (1980) - salience of onset makes it more likely to detect mispronunciation at start of word than end
- lexical search starts immediately
- Allopenna et al., (1998) - additional late actication - info at later points can still activate lexical selection
- increase in fixation probability for word that rhymes
Access to meaning (speech recognition)
Deciding which meaning of a word is correct
- Swinney (1979) - ambiguous vs disambiguous context - bug = ant or spying device
- for early target - both dominant + non-dominant meanings activated by the prime
- for later (200ms) target - dominant meaning is only one activated –> has had time to disambiguate
- different meanings initially activated but context helps select
Context effects (speech recognition)
Important for noisy environments
- Warren (2008) –> phoneme restoration effect
- if cough sound over one phoneme, still hear word intact
- Pollack & Pickett (1964) –> spoken word recognition decreases significantly if presented out of context + alone
- Tyler (1984) –> context effect in gating
- set size decrease happens faster in context
- faster recognition
Speech processing in the brain
Hickok & Poeppel (2007) - dual-stream model
- ventral stream –> maps sound to meaning
- for comprehension
- bilateral activity in Heschl’s gyrus, superior+ medial temporal areas
- Dorsal stream –> maps sound to articulation
- for more complex input like sentences
- left-lateralised, particularly left inferior frontal gyrus (Broca’s area)
Gil-da-costa et al., (2006) –> looked at sound-meaning mapping in evolutionary context
- rhesus macaques - exchange meaningful calls - stimulates comparable activity to humans
- PET –> activity in homologs of human language areas for salient vs non-salient stimuli
Rillig et al., (2008)
- left inferior frontal gyrus activation for complex input (sentences) –> dorsal stream
- temporal + frontal connection is unique to humans
- for grammatical + syntactic processes
- temporal lobe projection of human arcuate fasiculus is absent/smaller in non-human primates
- may be relevant for language evolution
Sentence processing
- outline:
- sentence
- syntactic rules
- syntactic parsing
- state main 4 theories of sentence processing
Sentences = not stored, need rapid mechanisms for parsing + extracting information
Syntactic rules - govern how words can be combined in a language (indepentend of meaning) –> abstract enough for ALL grammatically correct sequences BUT still rule out illegal grammar
4 main theories:
- Garden-Path model (Frazier & Rayner, 1982) –> initial parsing is purely syntactic, meaning informs later
- Constraint-based theories (MacDonald et al., 1994) –> initial interpretation depends on all available sources of information
- Unrestricted race model (Van Gompel et al., 2000) - all sources of info used to identify a syntactic structure
- ‘Good enough’ representations (Ferreiva et al., 2002) –> type of information used depends on the task - has to be good enough for context
(1+2 are the main ones)
Sentence parsing cues (state)
- syntactic principles
- statistical regularities
- grammatical categories
- prosodic cues
- semantic information
- world knowledge
Syntactic principles (sentence parsing)
Frazier (1987)
- late closure principle –> new items attached to the phrase/clause most recently processed
- optimises sentence comprehension
- decreases cognitive demand
- decreases STM requirements
- Minimal attachment –> links each incoming word to existing structure using simplest syntactic structure consistent with the input
- parsimony
Statistical regularities (sentence parsing)
Word order: english = SVO; japanese = SOV
Slobin (1966)
- presented participants with active/passive sentences, either reversible or irriversible (S+O swappable)
- RT matching sentence to pictures decreased for all active - because more fequently encountered
Grammatical categories (sentence parsing)
- individual words provide cues for sentence interpretation
- e.g. articles, prepositions, pronouns
- and = new phase node
- that/which = embedded sentence
Prosodic cues (sentence parsing)
Relative emphasis of different parts of the sentence
Beach (1991) - sentence completion task –> 2 types of sentence with same verb (‘argued’):
- Direct object - forcefully short vowel + stable pitch contour
- Complement - long vowel + fall-rise pitch
- if DO verb –> completed sentence in a way that had DO characteristics
- same result no matter how long the segment was
- prosodic cues define elements of the sentence and how they sit together
Semantic information (sentence parsing)
Meaning of individual elements
Trueswell et al., (1994) –> 2 sentences, one word changes
- reaction time longer (slower) if have to read 2 animate things in same sentence - creates more of a conflict in putting the sentence together
World knowledge (sentence parsing)
Effect of violation of world knowledge are rapid and parallel to those for semantic violations
Hagoort et al., (2004): ERP + fMRI - 3 sentences, 1 true, 1 world violation, 1 semantic violation
- world knowledge violation + semantic violation produce response at the same time (N400)
- semantic + world knowledge are equally fast + influence sentence processing in parallel
Neurobiology of syntax
Vigneau et al., (2011) - meta-analysis of 128 studies of syntactic processing
- extended frontotemporal processing netword (mainly on L)
- frontotemporal white matter connections (L) –> dorsal (arcuate fasiculus) + ventral (extreme capsule)
Rolheiser et al., (2011) - white matter tracks + syntactic comprehension
- worse performance with less white matter density (patients with white matter damage)
Neuro-cognitive model (for syntax)
Friederici, 2011
- functional distinction between dorsal + ventral pathways in L hemisphere
Friederici & Gierhan, 2013 (looking at functional brain imaging):
-
ventral pathway (aSTG) for simple syntax
- putting elements together
- the uncinate fascile and the inferior fronto-occipital fascile subserve semantic and basic syntactic processes
- the FC and the TC are connected via at least two pathways
- the uncinate fascile (UF)
- extreme capsule fiber system (ECFS)
- supporting local phrase structure building
- [also semantic processing]
-
dorsal pathway (pSTG) for complex syntax
- dependencies/embedding
- complex computations
- one pathway connects the temporal cortex and premotor cortex –> for speech repetition,
- another pathway connects the temporal cortex and posterior Broca’s area –> for complex syntactic processes
- the tract from the STG to Broca’s area is relevant for complex syntactic processes
- [also auditory –> motor mapping]
Evolutionary context (sentence processing)
Rillig et al., (2004) - evolutionary expansion of dorsal frontotemporal pathway in humans (L) –> involved in processing of complex sentences
Fitch & Hausel (2004) - non-human primates aren’t capable of mastering complex syntactic combinations
- both humans + monkeys can spot simple sequence violations (look/no-look measure)
- only humans can spot complex sequence violations
Bilingualism - history
- historical costs
- benefits
- costs
COSTS (historically):
- Saer (1924) - in rural context, IQ of bilinguals = 86 vs 96 of monolinguals
- BUT: other factors involved
- Jesperson (1922) - brain effort required for bilingualism –> reduces ability to learn other things
BENEFITS:
- Peal & Lambert (1962) - positive correlation between bilingualism and intelligence when other factors controlled
- EF –> ‘bilingual advantage’
- metalinguistic knowledge:
- Ianco-Worrall (1972) - can you call a dog ‘cow’ and a cow ‘dog’ if you were creating a language
- BL (English-Afrikaans) more likely to say yes
- by 7-9yrs - 0.59 BL (4-6yrs = .38)
- English ML = 0.08 (4-6yrs), 0.38 (7-9yrs)
- Ianco-Worrall (1972) - can you call a dog ‘cow’ and a cow ‘dog’ if you were creating a language
COSTS:
- Disadvantage in receptive vocabulary in preschool + early school years
- given word + have to find picture it relates to
- bilinguals have 2 possible labels - so relative usage of each is less
- Tip-of-the-tongue –> temporary inability to produce a known name or word
- Gollan et al., (2005) - TOT higher for bilinguals on objects (2 labels) but no difference on famous or personal names
Bilingualism and the brain
- activation for L1 and L2?
Indefrey (2006) - meta-analysis –> only 2/5 showed greater activation for L2 than L1
De Bleser et al (2003) - PET: dutch ML vs Dutch-French BL - latte acquisition (10yrs)
- cognates (same meaning similar sound in L1+L2)
- L2 has same activation as L1
- noncognates
- greater activation in frontal regions due to increased difficulty + increased cognitive capacity required
Rodriguez-Fornells et al (2005) - early acquisition: German, German-Spanish (L2 by 3yrs)
- used L2 for work/study
- activation pretty much the same in both
Indefrey (2006)
- chinese (L1) tested 3,6,9 months into a Dutch learning program (L2)
- after 6 months of exposure - identical regions in L frontal areas as in native speakers
Hierarchical model of bilingual acquisition
Kroll & Stewart (1994):
- Initial stage:
- meaning of L2 word accessed via L1 analogue mediator
- lexical links between L2 –> L1
- L1 concepts
- Later stage
- link directly from L2 –> meaning (direct access withough L1 involved)
- can translate between L1 + L2
- Final stage (proficient bilinguals)
- bidirectional relation between them all
Prediction:
- in intermediate stage - should be differences in L1–>2 translation vs L2–>L1 (semantic effect)
- Kroll & Stewart (1994): confirmed with English-Dutch bilinguals:
- slower to translate L1–>L2 than L2–>L1
- due to being mediated by meaning
- slower to translate L1–>L2 than L2–>L1
Bilingualism - competing for access (in parallel)
Green (1998) - suppression hypothesis –> both L1 + L2 constantly competing for access, have to suppress one
Spivey & Marian (1999) - presented with grid, have to follow instructions (Russian-English BLs)
- distractor present (sounds like a russian word –> phonological overlap)
- distractor shows some fixation even if instruction in english
- process in parallel - simultaneously
Meuter & Allport (1999) - code-switching paradigm
- digit colour indicates the language to name in
- RT in switching trials is higher (as expected)
- RT for L2 is higher than L1
- slower to switch from L2 –> L1 than L1 –> L2 (becuase better at suppressing L2)
- takes a lot of effort to suppress L1 in L2 trials so there is a spill-over effect (suppression hypothesis)
Bilingualism and executive functioning
- the ‘bilingual advantage’
Suggestion that EF of BL is better due to constant inhibition of non-target language
- Bialystok et al., (2004) - Simon task - overcoming conflicting cues to make a decision
- interference greater for ML
- reduced Simon effect + age effect for BL
- reduced WM cost for BL
- better prepared to cope with conflicting info?
- Craik et al., (2010) - Bilingual patients with Alzhiemers
- diagnosed 4.3yrs later
- onset of symptoms 5.1yrs later than ML
- languagr experience may contribute to cognitive reserve - compensating for effects of neuropathology
The bilingual advantage - evaluation
CRITIQUES (Paap et al., 2015;16):
- some replication attempts failed
- some possible methodological problems (SS, SES, cultural differences, immigrant status)
- publication bias (more likely to publish positive)
- chicken + egg problem:
- BL –> better EF or
- better EF –> aquiring L2/3/4
Beyond bilingual advantage:
- Olguin et al (2019) - learning/using multiple languages is environmental/cognitive demand
- brains adapt to environmental demands to maintain/optimise performance
- even if behavioural performance the same, maybe use different neural mechanisms for ML vs BL
- BL may trigger adaptive reconfiguration of the system - for optimal performance
Gestures (communication)
- types
- functions
TYPES:
- Beats = coordinated with speech prosody but no relation to meaning of it
- simple + repetitive
- Pointing = simple conventionalised movements referring to spatial locationrs or to bind objects to them
- Symbolic gesture = movement with conventionalised meaning
- ok, thumbs up, fingers crossed
- Lexical gestures = non-repetitive, change in form
- relate to meaning of accompanying speech
Iverson & Goldin-Meadow (1998) - congenitally blind people gesture as much as sighted people
FUNCTION
- communicative + informative
- facilitating word production
- Morrel-Samuels & Krauss (1992) - onset of gestures: 500ms-1s before words
- familiar words = closer gesture asynchrony + briefer - less planning/effort
- Rauscher et al., (1996) - gesture effects on speech fluency
- fake electrodes to immobilise hands
- no gesturing when describing spatial-related content = more disfluency
- Morrel-Samuels & Krauss (1992) - onset of gestures: 500ms-1s before words
Pauses and disfluencies (communication)
Breaks/irregularities/utterances not consistent with specific grammatical construction and occur within otherwise fluent speech (~6/100 words) - ‘uh’/’er’/’like’
Informative
- 60-70% of pauses fall at juncture of sentences
- disfluencies often at beginning of utterance (planning)
Brennan & Schober (2001): move mouse to square indicated in utterance
- between word interruptions
- move to the yellow purple square
- affects performance most
- move to the yellow purple square
- mid-word interruption
- move to the yel purple square
- mid-word interruption and filler
- move to the yel uh purple square
- speeds up comprehension
- move to the yel uh purple square
- disfluencies can signal speech repair - tell you to pay attention
Conversation convergence (communication)
- what is it?
- example
- reasons behind it
Temporary, flexible agreements between speaker and listener to view/name objects a certain way
Metzing & Brennan (2003): task = locate + move object in grid according to speaker’s instructions
- speaker names object, gives instruction + keeps that name for the rest of time - listener agrees
- NEW speaker:
- listener equally fast to detect target if new OR old word used (because no agreement with them)
- OLD speaker, NEW word:
- confusion and increased RT - because breaking social agreement
- Measured by eye movement - time looking at target
Garrod & Pickering (2004) –> speeds up communication
- coordinate - keep track of who says what
- distributes processing load between interlocuters
- maybe relies on simple priming + imitation mechanisms
Multimodal integration
- positives (speech/vision)
- preceding speech
POSITIVES:
- increases signal:noise ratio - raise up to 20dB
- Sumby & Pollack (1954)
- enhances comprehension of speaker if you see them - especially if complex speech/heavy accent
- Reisberg, McLean & Goldfield (1987)
Precedes speech:
- prelinguistic children link mouth movement to auditory signals
- Kuhl & Meltzoff (1982)
- macaques have highly deveoped audio-visual integration –> to support social interaction
- make similar eye movements to humans (focus on eye and mouth)
- single-cell recordings show cells in auditory cortex respond stronger to integrated signals
- Ghazanfar et al (2008)
Neural coupling (communications)
Neural coupling hypothesis: Hasson et al., (2012)
- stimulus-brain coupling –> to retrieve info about world + guide actions
- brain-to-brain coupling –> coordination among individuals in shared physical/social environment
Stephens et al (2010) –> temporal + spatial coupling of brain activity (speaker + listener in fMRI scanner)
- coupling in early auditory areas, superior temporal + inferior parietal gyri, insula + inferior frontal gyri
- synchronous speaker-listener coupling in early auditory cortices
- predictive anticipatory coupling in frontal lobe (PFC) - activation in listener precedes speaker
- level of comprehension correlates with degress of coupling in frontal areas
- neural coupling only in successful communication
Social contract in conversations - example (Tomasello)
Wyman, Rakoczy, and Tomasello, 2009
- “No, you can’t eat that. It’s soap!” (Wyman, Rakoczy, and Tomasello, 2009)
- Both understand and enforce the norms
- One block = soap
- Other block = sandwich
- Kids get mad at breaking the pretence
word segmentation
- phonological bootstrapping
- distributional learning
- statistical learning
-
Phonological bootstrapping
- Use phonotactic or prosodic cues (vowel length, pauses, loudness) to help break wall of speech
-
Distributional learning
- Track how often certain elements appear
-
Statistical learning
- How often certain elements cooccur with other elements
- Saffran, Aslin & Newport, 1996
- 8-month-old infants listened to 2 minute stream of syllables (made-up language – 4 x 3 syllable words)
- tibudogolatudaropipabikutibudodaropigolatudaropi tibudopabikudaropitibudopabikugolatu
- Then test words presented (2 types)
- Transitional probabilities: (between 1st + 2nd and 2nd + 3rd syllables)
- Test “words” from made-up language (pabiku, tibudo) = 1
- Only ever occurred together
- Test “non-words” not from language (tudaro, pigola) = 0.33
- From final syllable of one word + first 2 of another word
- String of syllables only occurred if 1st word followed by 2nd word
- Test “words” from made-up language (pabiku, tibudo) = 1
- During test phase – infants listened to non-words significantly longer than words could track transitional probabilities + distinguish between test words
- Infants learn structural properties of input
- Powerful learners
- Non-linguistic cognitive abilities play important role in language learning
- 8-month-old infants listened to 2 minute stream of syllables (made-up language – 4 x 3 syllable words)
- Once they break wall of sound, they can use distributional + statistical learning to analyse frequency + cooccurrence of certain patterns of sound/words/phrases
- Phonological bootstrapping to break up sound then distributional + statistical learning
Phoneme experiments
- categorical perception
- phonemic tuning
- head turn
- MEG
Eimas et al. 1971
- interested in how people perceive sounds on a continuum (e.g. b p)
- Do this by varying voice onset time (only way they differ)
- Sounds were evenly spaced across continuum but there was a noticeable point where b becomes p
- For adults voice onset time of +20 is heard as ‘ba’ and +40 is heard as ‘pa’
- +40 - +60 did not have same reaction
- Used high amplitude sucking technique to test on infants (1-4 months)
- Infants given special dummy to measure sucking rate
- Habituated to hearing ‘ba’ first until sucking rate gradually declined
- 6 stimuli used with VOT from -20 to +80 –> showed significant dishabituation in sucking rate (sped up) between +20 and +40
- Conclusion –> by age of just 1 years old they perceive sounds categorically in same way as adults
- Either /b/ or /p/ - cannot be both
Werker and Tees, 1984
- Young infants have categorical perceptions for sounds in other languages too
- Used pairs of sounds that sound the same to English adults but in other languages were contrastive phonemes
- Head turn procedure: Infant rewarded for turning head when sound changed
- English infants – declined until adult level
- Tuning in process –> when hear native language, infant becomes adept at discriminating sounds in native language
- In Japanese – infants born with r/l distinction not important in their language but after 1yr old they can’t tell them apart
- Specialisation –> restriction of options
Kuhl, Ramírez, Bosseler, Lin and Imada (2014)
- MEG to show that the ability to discriminate native versus non- native phonetic contrasts that emerges during the first year of life has both acoustic and motor components.
- tested English learning infants aged 7+ 11– 12 months, contrasting native English phonemes with native Spanish phonemes (via different voicing for / d/ or / t/ sounds).
- MMR showed that the younger infants could discriminate both native and non- native sounds,
- showed brain activation in both auditory and motor regions
- The MMR also showed that the older infants had lost the ability to discriminate the non-native speech sounds.
- double dissociation between acoustic and motor neural regions for the older infants (also present in adults)
- For auditory brain areas there was greater neural activation for the native speech sounds than for the non- native speech sounds
- for motor brain areas there was greater neural activation for the non-native speech sounds than for the native speech sounds.
- Suggested that infants’ increasing ability to form these speech sounds themselves could underpin this developmental dissociation
