Psychology 337

1
Psychology 337

• Section 6. Language Acquisition
• A. Pinkers Perspective
• (with additions)

2
First Words in English (Pinker)

• First words usually occur at about 1 year.
• For something to be a word, you must have some
  kind of concept attached to it.

3
Kinds of First Words

• 1. Almost 1/2 are words for objects
• Food, body parts, clothing, vehicles, toys,
  household items, animals and people.
• 2. Words for actions, motions and routines
• Up, off, open, peekaboo, eat, go.
• 3. Modifiers
• Hot, allgone, more dirty, cold.
• 4. Routines used in social interaction
• Yes, no, want, bye bye, hi, look at that, what is
  that.)

4
One Infants Early Words (Davis, l986)

• Word Times Used
• No 990
• Open 565
• Mommy 494
• Hi 451
• More 402
• Book 363
• Up 355
• Bye 311
• Baby 310
• Juice 303
• Doggy 296
• Uh oh 293
• Out 292
• Yeah 272
• Apple 244
• Cookie 240
• Cheese 236
• This 232

• Functional Categories of the Words
• 1. Social control (1) no
• 2. Food (5) juice, apple, cheese, cookie, bite
  (act)
• 3. Physical control (tool use) (5) open, up,
  out, more, off
• 4. Social names (3) mommy, baby, doggy
• 5. Greetings (2) hi, bye
• 6. Social crisis (1) uh oh
• 7. Plaything (1) book
• 8. Diexis (reference) (1) this
• 9. Social assent (1) yeah

5
Early Comprehension The Segmentation Problem

• The speech signal infants hear is usually
  acoustically continuous (not like writing, where
  there are spaces between words).
• So infants have to figure out where one word ends
  and the next one starts.

6
Motherese helps with Segmentation

• Mothers highlight key words by putting them on an
  acoustic pedestal
• Look at the BOTTLE
• Motherese also helps with emotional learning
• Approval/positive affect wide intonational
  range.
• Disapproval Low pitch, staccato.

7
Pinkers Idea About Segmentation

• First find key words.
• Then figure out residues between words.
• Even a dog might do some of this (from The Far
  Side cartoon)
• You say O.K. Ginger, Out of the garbage
• Dog hears Blah blah Ginger, Blah, blah, blah,
  blah, blah.

8
Segmentation Errors Occur

• Were going to Miami.
• I dont want to go to your ami.
• Behave
• I am hayv
• eaten by
• This apples been eatenby.
• The ants are my friend, theyre blowing in the
  wind.

9
The Specificity Hypothesis (Gopnik Meltzoff)

• What governs what words are first used?
• Specificity hypothesis
• Children acquire early words that are relevant to
  the specific cognitive problems that interest
  them.
• (Note Chomskian perspective does not include any
  link between cognitive and language.)

10
Two Early Concepts

• 1. Object Permanence.
• Finding objects after they are put out of view is
  related to the use of disappearance words e.g.
  gone.
• 2. Means-Ends Relationships.
• Using a stick to reach an object is related to
  the use of success/failure words e.g. there
  (Ive done it) or uh oh (failure).

11
Concept-Word Relations

• Gopnik and Meltzoff showed that object-
  permanence-related words are first used in the
  same week that a child reaches a certain level in
  solving object-permanence and means-ends
  problems.
• And this is not just a co-incidence (literally)
  of development some kids solve one type of
  problem before the other and the new words always
  relate to the problem type that has just been
  solved.

12
Beyond First Words

• 12-18 months 50 word stage.
• 2 things happen at about 18 months
• 1. Vocabulary spurt (New word every 2 hours).
• 2. First syntax (2 word stage)

13
Types of 2 Word Sentences (Slobin, 1979)pivot
open class
14
Grammatical Comprehension Precedes Production
(Huttenlocher)

• Kids at 1 1/2 years can understand which of the
  possible meanings that can be constructed from
  the following
• Give (or show) me (or mummy) the babys (your)
  bottle.

15
Output Bottleneck in Early Word Production

• (Roger Brown) Kids cant produce
• Mother gave John lunch in the kitchen.
• Even though they can produce all the components
  individually
• Mummy Fix.
• Give Doggy.
• Truck Window.

16
Next Stage All hell breaks loose.

• Between mid twos and late threes there is an
  output explosion. (Still ahead of researchers.)
• Syntactic types reach thousands before third
  birthday.

17
Key Development Embedding

• E.g. Im making it spin.
• Constituents
• 1. Its spinning.
• 2. Im causing it.

18
From Telegraphic to Non-Telegraphic

• Before three, speech is telegraphic (like
  telegrams), missing function words and
  inflexions.
• By the threes, function words are present more
  often than not.
• A full range of sentence types flower
  Wh questions, relative clauses, comparatives,
  negations, complements, conjunctions and
  passives.

19
Pinker Considers Correctness Amazing

• Sentences are grammatically correct about 90 of
  the time at age three.
• Auxiliaries 24 billion, billion possible
  combinations.
• Stromswald Virtually no errors in 66,000
  auxiliary constructions of preschoolers.
• All languages acquired with equal ease before the
  child turns 4.

20
Creative Mistakes

• A frequent past tense sequence. (Similar findings
  for the plural.)
• Time 1 went? Time 2 goed or wented? Time
  3 went
• 1. Irregular forms learned first (most frequent).
• 2. Then the more general regular form takes over.
• 3. Then the infant relearns the exceptions.

21
Another Error Type

• Example Dont giggle me.
• Causative rule takes an intransitive verb ,
  meaning to do something, and converts it into a
  transitive verb meaning to cause to do
  something.
• Correct forms
• Intransitive The butter melted.
• Transitive John Melted the Butter.
• But it doesnt work for giggle

22
Grammatical Competence with Functional
Incompetence

• The grammatical competence of three year olds
  contrasts sharply with their incompetence in many
  other realms
• E.g. Hitting a baseball, tying shoes.
• Pinkers conclusion It is plausible that the
  basic organization of grammar is wired into the
  subjects brain. (Chomskys position.)

23
Pinker Vs Chomsky on Natural Selection

• Unlike Chomsky, Pinker believes in natural
  selection.
• But unlike Bickerton he doesnt consider the
  question of how language might have evolved by
  natural selection he just assumes innate
  hard-wiring evolved.

24
Pinker on Nature-versus-Nurture

• Infants reared by animals or in closets dont
  talk.
• So experience is needed, in particular experience
  to speak language X.
• Meanings can be inferred from context.
• (Markmans whole object assumption.)
• Also Motherese helps.
• Slower, more exaggerated (more careful about
  making vowels clear than in speech to animals).
• More directed to the here and now.
• More grammatical (99.44 pure).

25
But Motherese is NOT

• Used in all cultures.
• Grammatically simple.
• E.g. of complexity of motherese To assemble the
  question What did he eat? from motherese He
  ate what requires a number of complex
  grammatical operations (Pinker, p 279).

26
More on Nature/Nurture

• Practice not essential. Many examples of good
  comprehension in mute people.
• But parents dont correct grammar, only
  meaningfulness (semantics).
• Result A kid who can bend meaning (lie) with
  good grammar.

27
What Experience Doesnt Give You

• Languages are infinite, childhood is finite.
  Therefore children cant memorize everything.
  They must leap into the linguistic unknown. But
  there are untold pitfalls Tense and number
  rules, causative rules
• She seems to be asleep? she seems asleep.
• She seems to be sleeping? she seems sleeping.
• Learning without negative feedback is a big
  argument for innateness.

28
Pinkers Solution

• Build in the basic organization of grammar.
• How does it work? These sentences give you
• Jane eats chicken Jane likes chicken.
• Jane eats fish
• Jane likes fish
• But then add This gives you (syntaxwise)
• Jane eats slowly Jane might slowly.
• Jane might fish Jane likes slowly.
• Jane might fish.

29
The Moral of this Example

• The child must couch words in grammatical
  categories like noun, verb or auxiliary.
• Meanings help in this
• Words for objects and people are nouns.
• Words for actions are verbs
• Words for paths and places are prepositions.
• Words for qualities are adjectives.

30
But Forming Class Knowledge is Not Enough

• Even if you know bother is a verb you cant
  figure out what word comes before it
• Noun That dog bothers me.
• Verb What she wears bothers me.
• Adjective Music that is too loud bothers me.
• Adverb Cheering too loudly bothers me.
• Youve got to understand complex noun phrases.

31
Solution?

• To put it crudely, the X-Bar theory of phrase
  structure must be innate. Also, since meaning of
  parents sentences is usually guessable in
  context, the child could use the meaning to set
  up the right phrase structure. (Example on P 285)
• Pinker also considers nouns and verbs to be
  innate categories.

32
Limiting the Possibilities

• Pinker considers that an innate capability serves
  to limit the analysis possibilities because
  without limits, there are simple too many
  possibilities to analyze.
• Alternative hypothesis The infant is
  self-limiting in general. Elman showed that his
  connectionist model couldnt learn grammar if you
  gave it whole sentences at a time, but could
  learn if you gave it only 3 words at a time.
  Maybe the infants input limitations are similar.

33
The Mapping Problem (Slobin, l979)

• Kids the same age have the same mental
  capabilities everywhere but they have to map
  cognition onto language in different ways in
  different languages.
• Daddy gave me the ball.
• ACTOR ACTION RECIPIENT (DEFINITE) OBJECT
• (Past)
• Der vater gab mir
  den ball.
• (DEFINITE) ACTOR ACTION RECIPIENT
  (DEFINITE) OBJECT
• (Sing, Masc, Subj) (Past, 3rd P, Sing)
  (Sing, Masc, Obj)
• Aba natan li et
  ha kadur.
• ACTOR ACTION RECIPIENT (OBJECT
  PARTICLE) (DEFINITE) OBJECT
• (Past, 3rd P, Sing, Masc)
• Babam bana topu verdi.
• ACTOR RECIPIENT OBJECT ACTION
• (Possessed by spkr) (Definite) (Past, 3rd
  P, Sing, witnessed by spkr)

34
Psychology 337

• Section 6. Language Acquisition
• B. Tomasellos Perspective
• (A Functionalist View)

35
Tomasellos Cut Paste Model

• This usage-based model has 3 propositions.
• 1. Utterance as a unit.
• Children hear and attempt to learn utterances,
  (i.e. phonological forms for expressing
  communicative intentions). In doing so they
  assign a communicative function to both the
  utterance as a whole and certain of its
  constituents.

36

• 2. Construction of utterance and constituent
  schemas with/without slots.
• Children store comprehended and produced
  utterances and constituents along with their
  functional analyses. Token and type frequencies
  of stored pieces of language lead to the gradual
  and piecemeal construction of (a.) utterance
  schemas and (b.) constituent schemas, both of
  which may contain relatively abstract slots.

37

• 3. Cutting and Pasting Schemas.
• Children produce utterances by cutting and
  pasting from their inventory of stored
  utterance schemas.

38
Main Proposition

• The basic unit of linguistic communication, not
  language, is the utterance.
• It is bounded by pauses and has a distinct
  intonation contour. It is observable in
  turn-taking in adult-child discourse.
• The child attempts to comprehend the overall
  communicative intention behind an utterance and
  behind the utterances individual constituents by
  blame analysis.

39

• Repeated occurrence in different contexts with
  functional consistency aids constituent
  identification
• Theres a ball. Gimmie my ball.
• The balls rolling. I want a ball.
• Throw the ball. (etc...)
• The process is to be distinguished from the
  purposeful presentation of words in the same
  context in Motherese
• Thats a ___
• Its a __ (etc...)
• Words are not learned in isolation or learned
  directly, but in the communicative context of the
  utterance, by determining their communicative
  sub-function in the utterance.

40
Creating Schemas

• Stored utterances that are similar in
  phonological form and communicative function
  congeal into Utterance Schemas.
• Utterance schemas
• More or less automated procedures for fluently
  expressing communicative intentions.

41
Utterance Schemas Cont.

• Limiting case A single invariant utterance.
• E.g There ya go. or I dunno.
• Other extreme Highly abstract schemas like the
  ditransitive NP VP NP NP.
• E.g. John put the book on the table.
• More normal case Phonological core with slot/s.
• E.g. Whats NP doing V-ing.

42
Development of Utterance Schemas

• First utterance schemas
• A consistent phonological form with a consistent
  communicative function. No slots.
• E.g. The first single words in the 50 word
  period. (Holophrases).
• Later
• Particular phonological forms with one open slot.
  (Pivot plus open class forms.)
• E.g. There ____. More ____
• These are isolated routines not interrelated or
  organized into a GRAMMAR.

43
Cutting and Pasting Schemas into Utterances.

• Usage-Based Syntactic Operations
• 3 options at this stage
• 1. Produce simple utterance schema
• Up. There ya go.
• 2. Retrieve and tweak.
• A. New constituent into slot Wheres the __
  ball.
• B. Tack on new constituent Throw it Here
• C. Insert constituent into middle (e.g.too)

44
Option 3.

• Combine constituent schemas without an utterance
  schema. E.g. skin ache.
• All this is not beginning with morphemes and
  gluing them together with rules. Rather the
  child starts with already constructed pieces of
  language of various shapes, sizes and degrees of
  abstraction.
• Not mindless but based on functional fit.

45
3 Subsequent Processes

• 1. Abstracting across utterance schemas to create
  more abstract constructions. E.g. Simple
  transitive, intransitive, passive.
• 2. Abstracting across constituent schemas to
  create more abstract constructions E.g. NPs, Ns,
  VPs, Vs, PPs, Ps, and perhaps some function word
  categories E.g. determiner.
• 3. Coordinating utterance schemas and constituent
  schemas in creative ways in complex constructions
  and discourse E.g. adverbials, coordinations.

46
Summary of Studies

• The utterances children hear.
• Cameron-Faulkner et al. Mothers speech to
  2-year olds.
• They hear 5-7,000 utterances per day!
• Only 15 had SVO form. Most utterances were
  questions (32), imperatives, (9), copulas
  (forms of the verb to be) (15), and sentence
  fragments (20).

47
Studies cont.

• More than 1/2 the utterances began with 1 of 52
  highly frequent item-based frames (more that 40
  times per day for more than 1/2 the children)
  consisting of 2 words or morphemes.
• 45 began with just 1 of 17 words, including
  What (8.6), That (5.3), It (4.2), You
  (3.1), Are/arent, (3.0), I (2.9).
• The rate of use was often highly correlated with
  adult use.
• Conclusion Most input highly repetitive forms.

48
Studies cont.

• The item-based nature of early language.
• Early words and phrases used in highly restricted
  syntactic contexts, proceeding gradually in
  piecemeal fashion.
• E.g. Pizzuto Castelli Of 6 possible
  person-number forms of the present tense verb in
  Italian, 1/2 of all the verbs took 1 form only,
  and 40 more took 2 or 3 other forms. Half of
  the verbs with more forms than this consisted of
  highly frequent, highly irregular forms that
  could have been learned by rote.

49
Studies cont.

• Item-based nature of early language continued.
• Tomasello and Brooks New verbs (e.g. tamming-
  like roll or spin) introduced in an intransitive
  construction, The sock is tamming, were seldom
  used transitively in appropriate test contexts,
  Hes tamming the car.
• Conclusion When learning about tamming they are
  just learning about tamming.
• Overgeneralization errors such as Dont fall me
  down are almost never produced before age 2 1/2
  to 3.

50
Studies cont.

• More complex constructions
• Dabrowska found that 83 of Wh question
  constructions during the 3rd year came from 1 of
  just 20 formulas, some correct (How did __) and
  some incorrect (Why I cant).
• Almost all individual Wh-Auxiliary pairs were
  either produced 100 correctly or 100
  incorrectly, showing their item-based nature.
• Conclusion Infants at 2-3 years of age mostly
  learn by imitation (This is denied by Chomsky).
  They dont yet have enough experience to
  construct adult-like linguistic abstractions.

51
Studies cont.

• The Process of Abstraction.
• Childers and Tomasello. Kids are better at using
  new verbs in transitive constructions if they
  heard them first in pronominal (Im V-ing it) as
  well as nominal contexts. How do they go from
  concrete to abstract?

TR DEICTIC SPEAKER
52
Tomasello, Fig. 1 Three levels of schemacity in
the childs Constructional schemas. TR
trajectory LM landmark.
TR DEICTIC SPEAKER
LM JUICE
WANT
I
want
juice
TR ANIMATE
LM THING
WANT
X
want
Y
TR ANIMATE
LM JUICE
PROCESS
Z
Z
Y
53
Studies cont.

• The Process of Constraint.
• More entrenched (more often heard and used) verbs
  are more resistant to overgeneralized syntactic
  use - e.g. transitive to intransitive and vice
  versa.

54
Studies cont.

• Usage-based Syntactic Operations.
• Lieven et al. tracked the history of the 300 plus
  different utterances produced on the last day of
  a 6-week study period.
• 2/3rds were things they had said previously.
• Most of the remaining third were repetitions of
  an established utterance schema plus other
  linguistic material filled in to a slot or
  added on to the beginning or end
• E.g. Wheres the ___ butter. Utterance
  schemas used on the last day had already been
  used 2-400 times. 3 utterances seemed to be
  novel combinations of utterance constituents.

55
Tomasello, Fig. 2 Two examples of cut-and-paste
operations. Utterance schemas are in italics,
constituent schemas are in regular type,
previous frequencies are in brackets,
functional characterizations are in CAPS, and the
arrows represent filling in (vertical) and
adding on (slanted) usage-based syntactic
operations.
TARGET UTTERANCE Lets roll it together HOW
PRODUCED Lets ACT it. 3 roll 5
together 4
TARGET UTTERANCE I want tissue lounge HOW
PRODUCED I want OBJ. 50 tissue 9
lounge 3
56
Studies cont.

• Complex Utterances.
• Sentential complements (Diesel Tomasello).
  Almost all these were formed from an existing
  sentence schema with one of a few
  complement-taking matrix verbs
• e.g. epistemic verbs such as think and know.
• Almost all kids just used think to indicate
  uncertainty and just in the first person
  singular. No other person forms or tense forms,
  and no negative forms.
• Also no complementizers used (I think that ).
  Attention-getting verbs (look, see) used
  almost entirely as imperatives.

57
Conclusion

• Utterance Schemas are big words.
• Their acquisition is like that of words - more or
  less rote attachment to functional contexts.
• Alternative (Chomsky-Pinker) Dual-Process theory.
  The linguistic world has two independent parts
• 1. All of the idiosyncratic aspects of language
  that must be learned in the normal way (the
  linguistic periphery).
• 2. All the regular aspects of language that dont
  need to be learned in the normal way because they
  are assimilated to a rule (The linguistic core
  that hooks up in some mysterious way with an
  hypothesized universal grammar).
• Basic choice 1 process or two.

58
Psychology 337

• Conclusions

59
Final Conclusions

• Basic Theme of the Course
• Two different approaches to the evolution of
  language.
• 1. Classicism (Plato, Descartes, Chomsky) The
  form of language is largely given in advance.
• 2. Neodarwinism Form arises from successful use
  (by natural selection).

60
Pinkers Stance

• Pinker tries to have his cake and eat it too.
• Form in advance is considered to result from
  natural selection.
• But he doesnt concern himself with how this form
  evolved by natural selection.
• To me, anyone who takes this kind of stance is a
  closet Platonist.

61
Bickertons Stance

• Bickerton does try to tell us how we got form by
  natural selection
• Argument structure results from reciprocal
  altruism.

62
Self Organization

• This remains an important way in which we can get
  complex patterns without prior specification.
• Note Lindbloms Elevator Principle for
  determining the vowel systems of languages.
• Also, I think the frame/content mode evolves and
  develops by self-organization.

63
One Conclusion

• Too much emphasis is presently being placed on
  genetics (in a way, its the new a priori form),
  even though the genes are so far removed from the
  final outcome of evolved language.
• This applies to Maynard-Smith and Szathmary, as
  well as to the modern descendents of the
  classical approach.

64
A Resultant Problem

• How do you get from genes to outcomes?
• And remember here what an important role is
  played by culture (e.g. memes for words), which
  in many ways must be remote from genes.

65
One Big Conclusion

• Language is definitely different from
  communication systems of other animals.
• The task is nevertheless to get to language via
  descent with modification, despite the huge
  difference, rather than postulate a hugely
  different cause.

66
The Importance of Productivity

• Of the 4 basic properties of language -
  Arbitrariness, Displacement, Duality and
  Productivity - productivity is perhaps the most
  important because it gives language its power.
  It allows an infinite set of sentences allowing
  us to talk about anything we can think about
  (Remember the analogy with genetics - the
  particulate principle.)
• Recursion is a big factor in languages
  productivity, but it might have come rather late
  in evolution.

67
Most Important? Words

• We had to invent words first, and that is the
  main thing that set us apart from other species.
  To have productivity at the syntactic level we
  had to have words. No recursion without
  cursion.

68
Bottom Line

• Darwins theory, basically accepted today,
  provides an unrivalled conceptual framework for
  approaching the explanation of any important
  phenomenon in biology.
• Language is no exception.

69
Speech

• I presented a possible Neodarwinian descent with
  modification scenario for the evolution of speech
  - The Frame/content theory.
• Descent sequence Ingestive mandibular
  oscillations to lipsmacks, to syllable frames, to
  programming the internal content of these frames.

70
Frame/Content Theory

• Two major claims
• 1. Ontogeny recapitulates phylogeny, except that
  ancestors made the model rather than learning
  from it.
• 2. Frames and content are separately represented
  in the brain in two systems also used for other
  actions The internal system (Frames, SMA) and
  the external system (Content, Brocas Area).

71
Hemispheric Specialization

• My Postural Origins Theory
• Left hemisphere speech-language arose from a left
  hemisphere specialization for whole-body control
  under routine circumstances. The right
  hemisphere is for emergency reactions.
• The main alternative view is anthropocentric
• Left hemisphere specialization arose first in
  hominids for tool use - speech/language. Right
  hemisphere specializations arose later by default.

72
Speech Perception

• The main material relevant to the overall theme
  of the course was a refutation of a claim of a
  priori form, the claim that there is innate
  speech-specific categorical perception.
• It turned out that other animals had it, and it
  is apparently a consequence of basic structural
  properties of the mammalian and avian auditory
  systems.

73
Language Comprehension

• Main points
• The difference between humans and computers
  regarding memory.
• The notion that you have to use abstract form
  class categories (noun, verb, etc) in order to
  understand language.

74
Language Acquisition

• We considered two approaches
• 1. Pinker Language must be innate, because it is
  so independent of input. Whatever sparse input
  there is cannot be organized without a guiding
  principle built in, a Chomskian universal
  grammar.
• 2. Tomasello If you look in greater detail than
  the Chomskians do, at day-by-day acquisition, you
  will see that it is extremely derivative of the
  input. Infants laboriously choose bits of input
  on the basis of their perceived communicative
  function, and achieve complexity by cutting and
  pasting bits of these input-based memes together.