Language acquisition Flashcards
Evidence for language specificity: brain modularity
Broca area:
-Appears to have something to do with grammatical processing
-Adjacent to the part of the motor control area for the jaws, lips, and tongue
-Damage to Broca’s area produces a certain kind of aphasia (language difficulty) resulting in stilted, ungrammatical (but contentful) speech
Wernicker area:
Appears to have something to do with meaning and word access
Adjacent to the primary auditory area that receives linguistic input
Damage to Wernicke’s area produces a certain kind of aphasia resulting in fluent speech that is completely lacking in sense
Counter to language specificty
Many non-linguistic things appear localised in the brain (e.g., expert piano players; sheep faces in shepherds)
Many aspects of language (such as word and concept meaning) appear to be spread throughout the entire cerebral cortex
What do feral children tell us about language learning
Strange gait (often all fours)
Odd senses (smell/hearing focus)
Poor social skills (eye contact, disinterest, little empathy)
Dislike of clothing
Vocabulary usually better than grammar, but sometimes no language at all
Confounding factor from observing feral children
Seriously deprived in many ways, not just linguistically
Very traumatised
How much do linguistic difficulties arise due to poor social development?
We know very little about their initial state or who they were
Highly variable environments
Studies on deaf children
- Performance on grammatical tasks in ASL (American Sign Language) depends highly on the age it was learned
- A similar study in BSL (British Sign Language) found a similar dependence on age: those who acquired it later did worse, even compared to second-language (L2) learners
Second language learners
Performance on grammatical tasks in one’s second language depends highly on the age it was learned
-idea of sensitive period for language learning
What are phonemes
Units of sound (consonants or vowels) in a language: the shortest segment of speech that distinguishes two words
Volwels
- Sounds where the air is not blocked
- Depend on the shape of the mouth
- Depends on where in the mouth the vowel is pronounced
Consonants
Sounds where the air is blocked
Depends on voicing: when the vocal cords begin to vibrate
Depends on place of articulation: where in the mouth the obstruction is
Depends on manner of articulation: how the blockage occurs
Cross-linguistic distribution of phoneme
Languages vary dramatically in total number of phonemes
But there are commonalities and patterns governing which phonemes are most frequent
Many distinctions found in one language do not occur in others
How is phonemes regconise
-many thought it was regconise in a continum but its actually catagorical
-In categorical perception, our perceptual system imposes a discrete category even though the underlying physical stimulus is continuous.
Consonants are perceived categorically, but vowels aren’t.
Testing Is categorical perception inborn, or does it develop through
experience with language? When does it emerge?
✤ Test via habituation: infants get bored if presented with the same stimulus for long enough
✤ We know they perceive two things as the “same” if they are habituated to the first thing and stay bored when presented with the second
✤ Habituate infants by playing the same phoneme over and over until they get bored
✤ Test: Present them with a new phoneme and see if they recognise it as “different” (between-category)
✤ Control: Present them with a new phoneme that differs by the same VOT, but does not cross a phonemic boundary (within-category)
Testing Is categorical perception inborn, or does it develop through
experience with language? When does it emerge? result
✤ So infants appear to have categorical perception from a very early age: 3 months or less
✤ As far as we can tell, they can perceive all consonants in all languages — even though adults cannot!
-it is likely an innate knowledge but not language specific as other animal can also achive this
Distinguishing phonemes after 12 monthj
Even though very young infants can hear all phonetic contrasts, by around 12 months of age they can only hear those that are in their native language
Developmental trajectory of language acqusition
-Birth Can perceive consonant contrasts in all languages categorically
6 months Babble native-language speech sounds
12 months Lose the ability to perceive consonant sounds that are not in one’s native language
Statistical learning:
Statistical learning: Sensitivity to the statistics of the environment: which things occur, how often they occur, and which things they occur with
Two general kinds
1. Individual: how often and in what distribution does a thing occur
2. Co-occurence: how often and in what distribution do two different things occur together
Phoneme learning = statistical learning test
Languages differ in the statistical distribution of their
phonemes. Could this be driving learning
Test: Train babies on the distribution of a new language, and see if they learn the new contrast 8-month-old English-learning babies, trained on a Hindi contrast that doesn’t occur in English
Control: Heard a series of tones
Unimodal: Distribution favoured one category
Bimodal: Distribution favoured two categories
Phoneme learning = statistical learning test result
Infants dishabituate only in the bimodal condition, suggesting they used that distribution to learn the Hindi phoneme contrast
Word segregation
the ability to separate sound in a sentence into coherent words (there is no pause when speaking)
Word segmentation: Possible solutions
-Perhaps people use phonotactic (recognize the pattern based on their knowledge) constraints: limitations on which sequences of sounds are permissible in that language
- Perhaps people use prosodic constraints influencing which stress patterns are common in that language
infants are aware of both of these by 9 months old
Intuition: words are “chunks” of language that always have the same sequence of phonemes. learn via statistical learning
Test for Word segmentation intuition in babay
habituate infants to a stream of speech whose “words” are defined solely based on TP(transition probability)s.
-there are 4 word, each word have 3 unique syllable and never put right after itself -> 100% for word and 1/3 for non word
If they respond differently to those words in isolation later, this is an indication they segmented them out successfully
Compare response to partial word (pago) and non-word (kuti).
They listened longer (indicating surprise) to the non-words
word segmentation test in adult
Test: teach four adults an artificial language with 1000 word types and 60,000 tokens (10 hours of speech, listened to while exercising)
Test immediately and after 1-2 months.
Segmentation was far above random or yoked controls (people who took the test without training)
test after 3 year can regconise some.
racking transition probability is not a language-specific skill
Visual sequences
Action sequences
Spatial organisation
reason why Word learning is hard
Saussure: the arbitrariness of sign as the form of a word tells you very little about its meaning
Quine: the problem of reference .The meaning of any word is logically underconstrained
Children are really good at learning words
✤ The first words are piecemeal, but at some point (usually between 14-24 months) there is a vocabulary spurt characterised by very rapid learning
✤ Individual variability is tremendous and not that meaningful at this age
Vocabulary growth varies with SES and environment due to amount of word they hear
How do children learn words meaning
✤ Shape bias (children prefer to categorise (most) nouns by shape)
✤ Mutual exclusivity (children generally assume items don’t have more than one label)
✤ Size principle(multiple examples are evidence for the smallest category that covers them)
✤ Social reasoning (Infants only learn labels if the speaker is looking at the objects)
✤ Shape bias
-May be learned based on statistical associations between words and features of the categories they pick out
Evidence for learning: teaching children new nouns cleanly organised by shape speeds up later word learning but teaching material noun does not work
Size principle
three organisation level :superordinate (animal), basic(dog), subordinate (shepard)
Social reasoning discrimintation
Children do not learn labels if the speaker has previously mislabeled other items
Parts of speech
These are things like noun, verb, etc. Different languages vary on what parts of speech they have and there is no fully agreed-upon classification scheme,although there are some basic similarities -seperate into open class and close class Parts of speech are associated with different roles in the sentence (subject, object, etc). These are often called arguments.
Open class
- easy to add new members
- carry much of the content
- produced earlier
- easier for 2nd language learners
Closed class
- hard to add new members
- carry much of the grammar
- produced later
- harder for 2nd language learners
default word order
Most languages have a default word order
English: primary word order is SVO (subject-verb-object) secondary word order is OVS
Word orders vary across languages, mostly SOV and SVO
The important of word order
Language is compositional: the meaning of a phrase or sentence is not just a mixture of the meaning of its words
-different word order can result in drastic different sentences
Children learn word order and grammar
Children learn word order early
Even the first sentences, though they drop words, usually have the right order for the words that are produced
The rest of grammar comes quickly. Lots of individual varibility but rapid expansion
Importance of verb
Most linguists agree that, regardless of the word order, verbs in every language are the “heads” of the sentence
They determine what the arguments are
They determine if arguments are optional or not
Because of their importance, a lot of grammar learning is actually about verb learning
How Do we learn verb arguments
We probably don’t learn just by brute mimicry (i.e., only using verbs with arguments we have seen them used before)
People (adults and children, in the lab and out) are willing to say verbs in argument structures they have never heard before: they are generalising beyond the input
People don’t just generalise arbitrarily, however
There are patterns for which verbs occur with which kind of argument structures
Possible solutions to the logical problem of language acquisition (how do children learn when they are wrong)
✤ Maybe children are told when they get things wrong? (i.e., perhaps they receive negative evidence)
✴ This probably doesn’t play much role: children appear not to receive much, or notice it when they do
Possible solutions to the logical problem of language acquisition
✤ Maybe they are sensitive to more subtle kinds of negative evidence, like rephrasings that statistically occur more often
✴ This does occur (statistically), although it’s arguable whether it is enough to entirely solve the problem (Chouinard & Clark, 2003)
✤ Maybe they use implicit negative evidence about which argument structures don’t appear
✴ This does also happen: it predicts n that over generalisation errors should be more common for infrequent verbs, which does happen (Ambridge et al., 2008)
Morphological rules
We also have to learn the morphological rules that apply to verbs
Morphological rules are rules governing how morphemes can be used and combined in a language
Morphemes are the smallest units that convey meaning
Includes tense (i.e., when the action took place), which overlaps somewhat with mood (i.e., whether it is certain to happen) and aspect (i.e., whether it is ongoing)
Two verb tenses in english
Regular: add +ed
Almost all new verbs are regular, which indicates that the past tense rule is productive and has some psychological reality
Irregular: mostly occur in clusters based on similarity of stem
These are also psychologically real: people are more confused about new verbs that sound like irregulars
Why are some verbs irregular?
Irregular verbs are the most frequent, which suggests an evolutionary story
If a verb is infrequent, it’s unlikely to be wellmemorised. That means people are more likely to use the regular +ed rule when they use it.
Only the frequent verbs will be so well memorised that they are impervious to regularisation over time.
How is verb morphology learned
Follows a general U-shaped curve
early: few verbs are used, but most of them have correct tense morphology
middle: past-tense rules often overgeneralised
end: eventually these mistakes are corrected
Revolves around whether there are two opposing mechanisms (rules vs memorisation) or whether it all emerges from statistical learning
learning by bigram
learning the rules of a language = learning words immediately follow other words
There are long-distance dependencies between words in sentences, which suggests there is some deep or hidden structure (probably related to the verb) linking them
But People probably don’t use just n-grams to capture grammatical rules as its impossible to track all long distance dependencies
o what does a sentence look like?
Most linguists agree that it has phrase structure: phrases nested in with one another in our underlying mental representation
The underlying depiction of a sentence’s phrase structure is called its parse tree
Syntactic ambiguity
Syntactic ambiguity arises from sentences with the same words but different meanings. This is because they have different parse trees.
It’s also interesting because it suggests that parse trees are psychologically real, and tells us something about how we process language.
So how do we process sentences
People actually do our best to figure the phrase structure out as we’re hearing the sentence
Sometimes this results in us getting misled into choosing the wrong parse tree, and then having to go back and reanalyse it from the beginning. These are called garden path sentences.
This implies that sentences with larger long-distance dependencies tend to be more difficult
sentences process in children
Developmentally: even children are remarkably good at figuring out language “as it comes” — a skill distinct from but correlated with knowledge of vocabulary