Onthenatureoflinguistic computations:complexity, - - PowerPoint PPT Presentation

On
the
nature
of
linguistic computations:
complexity, development,
and
evolution.

Luigi
Rizzi University
of
Siena

Structure
of
the
talk

I. Language
as
a
computational
system:

• Some
elements
of
linguistic
computations:
Structure
building,

movement,
interfaces,
locality.

• How
to
model
language
invariance
and
variation:
On
the
format

and
locus
of
parameters. II.
Language
variation
and
language
acquisition:

• On
the
early
acquisition
of
abstract
word‐order
properties:
an

experimental
result. III.
A
typology
of
Merge:
a
complexity
scale
and
its
implications
for language
acquisition,
adult
knowledge
of
language
and,
possibly, for
the
evolution
of
language.

The
unbounded
scope
of
the
human
linguistic capacities

• We
constantly
understand
and
produce
new

sentences,
combinations
of
words
that
we
have never
encountered
in
our
previous
linguistic experience

• …
and
still
we
find
them
familiar
and
usable.
• Galileo,
Descartes,
Humboldt,…

Elements
of
syntactic
computations: generative
models

• The
linguistic
capacities
can
be
modelled
as
the
possession
of

a
computing
machine
(Chomsky
1957),
consisting
of
at
least two
kinds
of
entities:

• ‐

Inventories,
lists
of
elements
stored
in
memory
(words,…)
• ‐

Computational
procedures,
putting
together
elements

drawn
from
the
inventories
to
form
higher
order
units (phrases,
sentences,…),
recursive.

• PHON









SYNTAX









SEM

Alternatives
to
a
computational
approach?

Could
it
be
that

we
just
memorize
fragments,
sequences
of
words
and retrieve
and
reuse
them? ‐ No:

we
clearly
have
the
capacity
to
go
beyond
what
we
hear
and generate
new
structures. Could
it
be
that
we
create
new
sentences
through
analogical generalization
from
memorized
fragments? ‐ This

statement
acquires
a
content
only
if
we
try
to
make
precise what
“analogical
generalization”
means,
thus
explaining
why certain
conceivable
“analogical
generalizations”
are
never considered
by
the
language
learner.

Recent
developments:
Inventories

• Inventories:
shift
of
emphasis
from
the

contentive
lexicon
(N,
V,
A,…)
to
the functional
lexicon
(D,
Aux,
C,
T,
Asp,…) Functional
elements: ‐


create
configurational
skeleta
for
the
insertion
of
contentive elements; ‐ trigger
the
fundamental
computational
processes; ‐ express
basic
parameters
of
variation; ‐ give
rise
to
complex
configurations,
studied
in
“cartographic” projects
(Rizzi
1997,
Cinque
1999,
etc).

Recent
developments:
elementary computations

• Computations:
shift
from
concrete,
construction‐
• riented
rules
(for
relatives,
questions,
passives,…)

to
more
abstract
computational
ingredients: ‐ Merge, ‐ Move, ‐ Spell‐out.

Merge
as
the
fundamental
recursive procedure

• Merge:

C 2 (1)






A










B












A











B 
where
C
=
A,

or
C
=
B:
the
selecting
element projects. 


































Chomsky
(1995,
2001)

A
derivation

[
meet
Bill
] [
can
[
meet
Bill
]] [
Mary
[
can
[
meet
Bill
]]] [
that
[
Mary
[
can
[
meet
Bill
]]]] [
said
[
that
[
Mary
[
can
[
meet
Bill
]]]]] [
has
[
said
[
that
[
Mary
[
can
[
meet
Bill
]]]]] [
John
[
has
[
said
[
that
[
Mary
[
can
[
meet
Bill
]]]]]

A
tree

(3)



















T 3 






N






 T 


John 3 




















T





















V has








 3 

































V




















C 































said









3 















































C


















T 













































that








3 



























































N


















T 

























































Mary

 3 















 T


















V 




































































can 3 




















































































V















N 



















































































meet











Bill

A
development:
the
cartography
of
syntactic
structures ‐
the
C
system

(Rizzi
1997,
2004)












Force
P 








3 




 3 










Force










TopP 









 3 





































3 


































Top












IntP 3 3 


































































Int














Foc
P 











































































3 





















































































3 















 Foc













ModP 3 











































































































3 








































































































Mod












FinP 


































































































































3 





































































































































3 Fin Clause

The
C‐system
of
Italian













Force
P 








3 




 3 










Force










TopP 











che 3 




















a
Gianni


3 



































Top














Foc
P 





















































3 











































QUESTO






3 















 Foc














ModP 3 domani



3 

























































































Mod












FinP 



















































































































3 






















































































































3 Fin











Clause 











































































































































gli
dovreste
dire Credo
che
a
Gianni,
QUESTO
domani
gli
dovreste
dire ‘I
believe
that
to
Gianni,
THIS
tomorrow
you
should
say’

Movement

• Expressions
are
often
pronounced
in
a
position
different
from
the

position
in
which
they
are
interpreted. (6)a





Which
book
should
I

read
___
? 





b




Which
book
do
you
think
I
should
read
___
? 





c



Which
book
do
you
think
…
the
professor
said
…
we 











should
read
___
? (7)a



[

___
C

[

I
should

read
[which
book
]
]
]







MOVE
 




b


[
[
which
book]


should+C

[
you

___
read
___

]
]

Movement
as
Search+Merge:
Internal Merge

(1)


[
Q


[the
boy]

[saw
[which
girl]]]















Search

 (2)[
Q


[the
boy]

[saw
[which
girl]
]]















Internal
Merge
 (3)[
[Which
girl]
[
Q

[the
boy]

[
saw
<[which
girl]>
]]] Internal
and
External
Merge
are
dedicated
to
the
expression
of
two types
of
semantic
properties: ‐ External
Merge
expresses
(among
other
things)
argumental semantics
(who
does
what
to
whom). ‐ Internal
Merge
expresses
(among
other
things)
Scope‐Discourse semantics
(Scope
of
operators,
topicality,
focus,
etc.).

Division
of
labor
of
External
and
Internal
Merge, and
two
kinds
of
interpretive
properties
at
the interfaces
with
semantics
and
pragmatics

(1) [I
[will
[read
[your
book]
tomorrow]]]














Ext.
Merge
 (2)


Top
[I
[will
[read
[your
book]
tomorrow]]]





Search

 (3)


Top
[I
[will
[read
[your
book]
tomorrow]]]





Int.
Merge
 (4)


[your
book]
Top
[I
[will
[read
<[your
book]>
tomorrow]]] The
expression
your
book
is
merged

in
two
positions,
where
it
picks up
the
interpretive
properties
of
“patient
of
read”
and
“topic”, respectively.

So,
External
and
Internal
Merge
are
not
just
rules
of a
formal
game,
but
play
a
critical
role
at
the
interfaces
with semantics
and
pragmatics.

Division
of
labor
of
External
and
Internal
Merge, and
two
kinds
of
interpretive
properties
at
the interfaces
with
semantics
and
pragmatics

(1) [I
[will
[read
[your
book]
tomorrow]]]














Ext.
Merge
 (2)


Foc
[I
[will
[read
[your
book]
tomorrow]]]





Search

 (3)


Foc
[I
[will
[read
[your
book]
tomorrow]]]





Int.
Merge
 (4)

[your
book]
Foc
[I
[will
[read
<[your
book]>
tomorrow]]]


Ext.M (5)

It
is

[your
book]
Foc
[I
[will
[read
<[your
book]>
tomorrow]]]

Languages
which
overtly
express
Scope‐ discourse
heads

(5)a


Ik

weet

niet


[
wie

of

[
Jan

___
gezien
heeft
]] 







‘I


know

not




who

Q



Jan










seen

has’ 





























































(Dutch
varieties,
Haegeman
1996) 




b

Un

sè







[
do




[

dan




lo






yà




[
Kofi




hu




ì
]]] 








‘I



heard




that






snake
the



Top




Kofi


killed

it’ 





























































(Gungbe,
Aboh
2004) 




c

Un

sè







[
do




[

dan




lo






wè




[

Kofi



hu


___
]]] 







‘I



heard




that






snake
the



Foc





Kofi


killed





’ 





























































(Gungbe,
Aboh
2004)

The
interface
with
the
sound
system:
Pitch contour
of
topic‐comment
and
focus‐ presupposition
in
Italian
(Bocci
2009)

H+ +L* L-L% H+ +L* H+ +L* H+ +L* H+ +L* L-L% a mi he lan d_ elo d_ er ma ni ho vo r beprezen ta re pje ran d_ ela A Michelangelo Germanico vorrebbe presentare Pierangela 100 450 200 300 400 Time (s) 3.7942

L+ +H* L- L* L* L* L-L% a mi he lan d_ e lo d_ er ma niho vo rbbeprezen ta repje ran d_ e la A MICHELANGELO Germanico vorrebbe presentare Pierangela 100 550 200 300 400 500 Time (s) 3.61823

Locality

(1) How
do
you
think
[
he
behaved
___
] (2) *
How

do
you
wonder
[
who
behaved
___
] Relativized
Minimality:
in 


















…
X
…
Z
…
Y
… A
local
relation
cannot
hold
between
X
and
Y
across
an intervener
Z
if
Z
is
of
the
same
type
as
X.
(Rizzi
1990,
2004) (3) *How

do
you
wonder
[
who



behaved




___


] 
















































* 











X






































Z





























Y

Spell‐out
and
the
interface
with sound

Certain
positions
are
pronounced,
hence
at
the
PF
interface
they activate
instructions
for
the
articulatory
system. Other
positions
are

visible
on
the
LF
interface,
but
they
are
not spelled
out
on
the
PF
interface: (1)a


Johni
promised

Billk



[
PROi
to
shave
himselfi
] 





b

Johni
persuaded
Billk


[
PROk
to
shave
himselfk
] (2)


Which
picture
of
himselfi

did
Johni
choose

___
? (3)

Which
picture
of
himselfi

did
Johni
choose
<which
picture
of himselfi>

Language
invariance
and
variation

Comparative
linguistic
studies
show
that
some
properties
vary across
languages,
while
other
properties
are
invariant. Within
the
tradition
of
generative
grammar,
invariance
and variation
is
expressed
by
two
theoretical
objects: ‐ Particular
grammars,
which
expresses
the
properties
specific to
a
particular
language,
and ‐ Universal
grammar,
which
expresses
the
universal architecture
of
language.

Language
invariance
and
variation:
principles and
parameters
of
Universal
grammar

• Universal
grammar:
a
system
defining
the
general
architecture
common

to
all
human
languages,
specifying
a
set
of
universal
principles
and
leaving

• pen
a
set
of
choice
points,
the
parameters.
• Particular
grammar:
UG
with
parameters
set
in
a
particular
way.






























































































(Chomsky
1981) (inspired
by
work
in
fundamental
biology:
Monod,
Jacob,
etc.)

• Parametric
models
introduced:











‐

A
powerful
technical
language
for
doing
comparative
syntax; 









‐

A
viable
model
of
the
acquisition
of
syntax:
obscure
ideas
on
“rule induction”
were
replaced
by
a
clear
operation
of
selection
of
parametric values.

On
the
format
and
locus
of
parameters

Format
of
parameters
is: (1)


“H
has
F” where
H
is
a
functional
head
and
F
is
a
feature
acting
as
an instruction
for
a
particular
syntactic
action:
Merge,
Move, Spell‐out. Locus:
Parameters
,
initially
conceived
as
specifications
on principles,
are
more
naturally
expressed
as
specifications

in the
functional
lexicon.

A
typology
of
parameters

• Merge
parameters:
 ‐
what
category
does
H
select?


















 















‐
to
the
left
or
to
the
right?

• Move
parameters:



‐
does
H
attract
a
lower
head?












































‐
does
H
attract
a
phrase
to
its
Spec?

• Spell‐out
parameters:


‐
is
H
overt
or
null?

















































‐
does
H
license
a
null
dependent?

On
the
format
and
locus
for
parameters

(21)
Current
assumptions
on
the
numerosity
of
parameters
are
sometimes taken
as
a
kind
of
reductio
ad
absurdum
of
the
very
idea
of
parametric syntax,
the
idea
that
syntactic
diversity
is
amenable
to
a
finite
set
of binary
options;
so,
the
current
conception
is
sometimes
seen
as
an undeclared
retreat
to
the
EST
conception
of
grammars
as
systems
of language‐specific
rules. (22)
But
there
is
a
confusion
here
between
the
locus
and
the
format
of parameters:
under
the
current
conception,
the
loci
of
parameters
are quite
numerous
and
diverse,
but
the
format
is
extremely
restrictive,

as determined
by
the
restrictiveness
of
minimalist
syntax.
So,
the
parametric space
is
radically
more
restricted
than
the
space
of
possible
language‐ specific
rules
in
EST
models.

Parameters
reduce
to
triggering
the
basic syntactic
actions
of
Merge,
Move
and
Spell‐out.

A
merge
parameter

All
human
languages
are
merge‐based,
but
there
is
a
fundamental

• rdering
parameter:

(1)A
head
precedes/follows
the
complement. NB:
This
also
concerns
v,
n,
etc.,
the
functional
heads
which
assign
a categorial
status
to
a
lexical
root. Coherent
head‐initial
and
head‐final
languages: (2)
a



John
has
said
[
that
Mary
can
meet
Bill
] 





b


John‐wa



[Mary‐ga




Bill‐ni




a

‐

eru‐




to




]
itte‐aru 










‘John‐Top
[Mary‐Nom
Bill‐Dat
meet‐can‐


that



]
said‐has

VO
and
OV
Languages

T T N John T has V said V C that N Mary V meet N Bill C T V T can T

(1) (2)

T T N John-wa T

• aru

V itte- V C to N Mary-ga V a- N Bill-ni C T V T

• eru-

T

Interim
summary

Some
properties
of
language
design
that
the
study
of
the evolution
of
language
should
deal
with: ‐ Inventories:

the
contentive
and
functional
lexicon; ‐ Computations:

structure‐building ‐ Computations:
Movement ‐ Principles
at
the
interfaces
with
sound
and
meaning ‐ Locality
principles ‐ Invariance
and
variation

• How
are
these
properties
acquired?
• How
are
they
implemented
in
the
brain?
• How
did
they
arise
in
evolution?

The
acquisition
of
word
order properties

• The
parametric
approach

expects
that
the
acquisition
of
language‐

specific
properties
should
be
fast:

the
problem
is
well‐defined (selecting
one
of
two
possible
choices)
and
the
child
is
guided
by domain‐specific
cognitive
resources.

(Wexler’s
VEPS)

• In
fact,
early
productions

in
the
two
words
stage
already
conform

to
the
basic
word
order
of
the
target
language,
VO
or
OV,
as
VEPS predicts
(Poeppel
&
Wexler
1993).

• But
it
could
be
that
at
this
early
stage
the
child
simply
reproduces

fragments
that
he
hears,
and
that
abstract
grammatical
knowledge is
developed
much
later
(as
in
“constructivist”
approaches).

• SO,
the
important
question
arises:
how
early

does
the
child

possess
abstract
grammatical
knowledge?

J.
Franck,
S.
Millotte,
A.
Posada,
L.
Rizzi
(2010)
Abstract knowledge
of
word
order
by
19
months:
An
eye‐ tracking
study.

Subjects:
19
months
old
infants

exposed
to
French The
experiment
combines: ‐ The
“preferential
looking”
paradigm; ‐ Pseudo‐verbs
(jabberwocky)
. ‐ The
“weird
word
order”
paradigm: 
Grammatical
NP
V
NP

sentences: 



















(1)

Le

lion
poune
le
cheval 

Ungrammatical
NP
NP
V
sentences:




















(2)

L’âne
le
chien
dase

Preferential
looking
paradygm

(Hirtsh‐Pasek
&
Golinkoff
1996)

« Hey, she’s kissing the keys! »

Vidéo Match Vidéo Mismatch

Condi tion Audio sentence Video 1 Transitive action Video 2 Intransitive action NVN

Le lion poune le cheval The lion is pouning the horse

NNV

L’ âne le chien dase The donkey the dog is dasing

Results

J.
Franck,
S.
Millotte,
A.
Posada,
L.
Rizzi
(2009)
Abstract
knowledge
of
word

• rder
by
19
months:
An
eye‐tracking
study.

1000 2000 3000 4000 5000 6000 7000 8000 9000 Grammatical Ungrammatical Transitive action Intransitive action

Results

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

0-4 sec 4-8 sec 8-12 sec 12-16 sec 16-20 sec Proportion of looking to transitive action

NVN NNV

Conclusions

Infants
at
19
months
exposed
to
French
clearly
interpret
NP
V
NP structures
as
Agent
–
Action
–
Patient
with
novel
verbs,
while
they have
no
preference
for
the
interpretation
of
NP
NP
V
structures. They
thus
appear
to
have
abstract
knowledge
that
the
language
they are
exposed
to
is
SVO. If
the
infant
had
no
abstract
knowledge
(“my
language
is
SVO”)
and was
simply
memorizing
new
verbs
with
their
syntactic environment,
no
selective
preference
for
the
transitive interpretation
with
NP
V
NP
would
be
expected:
the
infant
would just
memorize
the
new
verb
with
its
idiosyncratic
environment,
and would
have
no
basis
to
prefer
the
transitive
scene
with
VO,
but
not with
OV.

Cues
for
the
early
acquisition
of
word

• rder

‐
Christophe,
A.,
Nespor,
M.,
Guasti,
M.T.,
&
Van
Ooyen,
B. (2003)

showed
that
babies
at
3
months
are
sensitive
to differences
in
prosodic

prominence
correlated
to
OV‐VO (French‐Turkish). ‐
Gervain,
J.,
M.
Nespor
,
R.
Mazuka,
R.
Horie,
J.Mehler
(2008) shows
that
at
10
months
infants
exposed
to
Italian
prefer
the

• rder
[
frequent
Word
–
infrequent
Word
]

in
an
artificial

language,
while
infants
exposed
to
Japanese
have
the

• pposite
preference.

An
interpretation

A
reasonable
conjecture
is
that
children
fix
the
ordering parameter
prelexically
in
the
first
year
of
life,
on
the
basis
of prosodic
and
statistic
cues. Once
lexical
meanings
of
N’s
and
V’s
start
being
systematically associated
to
segmented
words,

children
exposed
to
French correctly
interpret
SVO
structures
with
novel
verbs
on
the basis
of: ‐ The
abstract
parametric
knowledge
that
French
is
SVO; ‐ General
principles
determining
the
form
–
meaning
mapping: a
subject‐verb‐object
sentence
is
interpreted
as
expressing
a biargumental
action,
with

the
subject

interpreted
as
the agent,
the
object
as
the
patient.

“Language
faculty”
vs
“Language
as
culture”

Q:

How
is
it
possible
that
every
child

manages
to
acquire
a
natural
language, an
extraordinarily
complex
system? ‐

“Language
faculty”
approaches:
There
is
a

human
language
faculty

which severely
limits
the
class
of
possible
human
languages.

The
child
analyzes incoming
data
on
the
basis
of
dedicated
cognitive
resources
and
quickly selects
a
grammatical
system
on
the
basis
of
experience. ‐
“Language
as
culture”
approaches:

The
acquisition
of
a
language
is
like
the acquisition
of
culture,
or
technology.
Languages
can
vary
indefinitely: there
are
certain
contents
to
express,
there
are
indefinitely
many imaginable
ways
to
do
so,
linguistic
communities
make
particular
choices and
language
learners
figure
out
what
these
choices
are
as
in
any
other case
of
cultural
acquisition,
through
their
general
intelligence
and problem‐solving
capacity

(a
wide
spectrum
of
approaches,
e.g.,
Evans
& Levinson
2009,
BBS).

“An
instinctive
tendency
to
speak”

• [Language]
certainly
is
not
a

true
instinct,
as
every language
has
to
be
learnt.

• It
differs,
however,
widely

from
all
ordinary
arts,
for man
has
an
instinctive tendency
to
speak,
as
we see
in
the
babble
of
our young
children,
while
no child
has
an
instinctive tendency
to
brew,
bake,
or write… 
















(C.
Darwin
DM1871)

Comparing
the
two
approaches

The
problem
of
comparing
the
predictions
of
the
“language
faculty”
and
of the
“language
as
culture”
approaches
can
be
attacked
from
many
angles: ‐ Through
comparative
linguistics
(Baker
2001):
is
it
the
case
that
“anything goes”
in
language? ‐ By
assessing
“poverty
of
stimulus”
arguments
(Chomsky) ‐ Through
brain
imaging
techniques
(Moro
et
al.
2003):
how
does
the
brain process
“possible
and
impossible”
linguistic
regularities? ‐ Through
the
experimental
study
of
the
initial
cognitive
state
of
the newborn
(Mehler
et
al.) ‐ Through
the
study
of
development:
how
early
does
the
child
manifest knowledge
of
universal
properties?
And
of
language
specific
properties?

Three
kinds
of
Merge

Merge
is
usually
considered
a
unitary
formal
operation,
but
a more
refined
analysis
shows
that
there
is
a
typology
of
cases, which
can
be
naturally
ranged
along
a
dimension
of
increasing complexity. Primary
Merge:










[the
girl] Recursive
Merge:







[saw
[the
girl]] Phrasal
Merge:











[
[the
boy]
[saw
[the
girl]]
] Item
in
red:
selector Item
in
green:
selectee

Primary
Merge

Primary
Merge

(or
H‐H
Merge)
takes
two
items
from
the functional
or
substantive
lexicon
and
forms
minimal
phrases: [the
girl] [with
Mary] [see
people] Operative
memory:
the
lexical
array,
the
array
of
words
selected from
the
lexicon
at
the
beginning
of
each
syntactic computation. Non‐recursive.

Recursive
Merge

Recursive
Merge
(or
Head‐Phrase
Merge):
takes
an
item
from
the
functional

• r
substantive
lexicon
and
a
phrase
already
formed:

[the
[tall
boy]
] [see
[the
girl]
] Operative
memory: I. the
lexical
array
and

• II. a
memory
buffer
in
which
the
structure
already
built
is
stored.

This
is
the
fundamental
recursive
step
in
syntactic
computations. The
head
selecting
the
phrase
is
the
head
of
the
construction
and
assigns
its label
to
the
new
entity
created
by
Merge.

Phrasal
Merge

Phrasal
(or
Phrase‐Phrase
Merge)

takes
two
complex
phrases
already
formed by
Merge: (1)





[
[
the
boy
]

[
saw

[
the
girl
]
]
] Operative
memory: I. The
lexical
array
(because
selectional
properties
must
be
“seen”).

• II. A
primary
memory
buffer
(containing
an
already
formed
phrase
with
an

active
“selector”,
like
the
verb
in
(1)):



[
saw

[
the
girl
]]

• III. A
secondary
memory
buffer
containing
an
already
formed
phrase:

[
the

boy] Recursive?
The
point
is
controversial. The
label
of
the
construction
is
determined
by
the
selector.

Movement:
Search
+
Phrasal Merge

(1)


[
Q


[the
boy]

[saw
[which
girl]]]















Search

 (2) [
Q


[the
boy]

[saw
[which
girl]
]]














Phrasal
Merge
 (3) [
[Which
girl]
[
Q

[the
boy]

[
saw
___
]]] A
search
operation
conducted
within
the
primary
buffer
identifies
a
candidate for
Phrasal
Merge.
The
identified
candidate
is

then
merged
with
the whole
structure. Phrasal
movement
is
a
composite
operation
combining
a
Search
procedure with

the
possibility
of
phrasal
Merge.

Phrasal
Merge

• Phrase‐Phrase
merge
is

more
complex
than
Head‐Phrase
merge

because
it
requires
holding
in
operative

memory
two
structures computed
independently
and

potentially
quite
complex.

• An
indirect
sign
of
the
accrued
complexity
of
Phrase‐Phrase
Merge

may
be
the
reluctance
that
young
children
show
in
naturally producing
Subject
–
Predicate
structures
with
complex
subjects, and
the
tendency
to
drop
determiners
more
in
subject
than
in

• bject
position
(Guasti
et
al.
2005).

(1)a

The


boy


saw


the


girl 





b

___

boy


saw


the


girl













More
likely 





c

The

boy



saw


___

girl













Less
likely

Phrasal
Merge

Another
possible
indirect
sign
of
the
complexity
of
phrasal
Merge
is the
fact
that
in
adult
systems
phrasal
specifiers
tend
to
be “islands”,
while
phrasal
complements
are
not: (1)a


John
bought
[a
picture
of
Mary] 




b


Who
did
John
buy
[a
picture
of
___
]? (2)a

[A
picture
of
Mary]
pleased
John 




b
*
Who
did
[a
picture
of
___
]
please
John? Apparently,
the
content
of
the
secondary
buffer
is
treated
as
a
unit, and
is
not
accessible
to
further
syntactic
operations
(Uriagereka 1998)
,
while
the
content
of
the
primary
buffer
remains
accessible.

A
Merge‐
based
complexity
scale

1.
Selection
from
the
lexicon:
utterances
with
single
words. 







(certain
animal
communication
systems;
fragments
of
languages taught
to
apes?
one‐word
stage
in
language
acquisition?) 2.

Primary
Merge:
2‐word
utterances. 







(2‐word
stage
in
acquisition?
Predominant
one‐
or
two‐sign structures
in
home‐sign
systems,
with
very
few
multiple
signs?) 3.

Recursive
Merge:
generates
an
unbounded
set
of
utterances
of
the form
Head
–
Phrase. 






(possibly,
a
stage
reflected
in
the
reluctance
that
young
children show
for
the
production
of

complex
specifiers) 4.

Phrasal
Merge:
permits
an
unbounded
set
of
utterances
with complex
specifiers,
and
creates
the
option
of
phrasal
movement.

Implications
for
the
evolution
of
language?

‐ This
typology

also
invites
the
speculation
of
4
possible
successive steps
in
the
evolution
of
syntactic
computations:

• 1. Simple
access
to
the
lexicon,
leading
to
one
word
utterances;
• 2. Primary
Merge,
leading
to
at
most
two‐word
utterances;
• 3. Recursive
Merge,
leading
to
a
potential
infinity
of
Head‐Phrase

utterances

• 4. Phrasal
Merge,
leading
to
a
potential
infinity
of
phrases
with

complex
specifiers. 1
seems
to
be
commonly
attested
in
animal
communication
systems. 2
represents
the
first
rudimentary
step
toward
a
combinatorial system;
its
attestation
in
communication
systems
of
non‐human primates
is
highly
controversial,
so
it
seems
to
be
at
the
border
of the
computational
capacities
of
such
species.

Recursive
Merge

‐ The
critical
evolutionary
step
appears
to
be
3:
recursive
Merge, which
permits
mastery
of
a
potential
infinity
of
structures,
possibly resulting
from
a
“minor
reorganisation
of
the
brain”
(Hauser, Chomsky,
Fitch)
.
And/or
a
computational
capacity
extended
to language
from
some
other
cognitive
system
where
it
pre‐existed (recursive
“theory
of
mind”?). ‐ Of
course
the
emergence
of
such
a
computational
ability
could
not take
place
in
the
vacuum:
Merge
could
only
be
used
if
the
system already
included
a
lexicon,
a
list
of
elements
for
the
recursive computations
to
operate
on,
and
perhaps
the
intermediate capacity
to
combine
lexical
items
non‐recursively
in
pairs
(primary Merge).

Phrasal
merge

‐ The
extra
step
(4)
of
permitting
Phrasal
Merge
is
presumably linked
to
the
necessity
of
expressing
events
with
more
than

• ne
participant,
in
structures
in
which
each
participant
may

be
referred
to
by
a
complex
phrase. ‐ This
yield
the
possibility
of
producing
structures
with
complex Specifiers… ‐ …
an
option
which
in
turn
opens
the
possibility

for
phrasal movement,
an
operation

which
plays
a
critical
role
at
the interpretive
interfaces,
and
is
systematically
found
across languages.

Conclusion

A
study
of
the
evolution
of
language
cannot
be
conducted without
an
in‐depth
understanding
of
the
basic
ingredients and
functioning
of
linguistic
computations.
I
have
tried
to provide
an
overview
of
such
computations
according
to current
formal
models. ‐
The
rapidity
of
language
acquisition
and
the
early
appearance

• f
abstract
syntactic
knowledge
suggests
that
there
is
an

“instinctive
tendency
to
speak”
in
our
species
which
calls
for an
evolutionary
explanation.

Conclusion


A
Merge‐based
complexity
scale
seems
to
have
a
potential

explanatory
value
for
the
study
of
language
development
in normal
and
special
circumstances,
as
well
as
of
adult
linguistic computations. It
may
thus
help
us
to
identify
more
rudimentary
precursors
of modern
language
syntax,
which
may
still
be
manifested
in vestigial
forms

in
language
acquisition
and
language
creation in
special
circumstances.

















Thank
you!

References

Christophe,
A.,
Nespor,
M.,
Guasti,
M.T.,
&
Van
Ooyen,
B.
(2003).
Prosodic
structure and
syntactic
acquisition:
the
case
of
the
head‐direction
parameter. Developmental
Science,
6(2),
211‐220. Gerken,
L.‐A.,
Jusczyk,
P.W.,
&
Mandel,
D.
(1994).
When
prosody
fails
to
cue
syntactic structure:
9
months
olds’
sensitivity
to
phonological
versus
syntactic
phrases. Cognition,
51,
237‐265. Gertner,
Y.,
Fisher,
C.,
&
Eisengart,
J.
(2003).
Learning
words
and
rules.
Psychological Science,
17(8),
684‐691. Gervain,
J.,
M.
Nespor
,
R.
Mazuka,
R.
Horie,
J.Mehler
(2008)
Bootstrapping
word

• rder
in
prelexical
infants:
A
Japanese–Italian
cross‐linguistic
study,
Cognitive

Psychology,
2008. Hirsh‐Pasek,
K.R.,
&
Golinkoff,
R.M.
(1996).

The
origins
of
grammar.
Cambridge,
MIT Press. Matthews,
D.,
Lieven,
E.,
Theakson,A.,
&
Tomasello,
M.
(2007).
French
chilldren
use and
correction
of
weird
word
orders:
A
constructivist
account.
J.
Child
Language, 34,
381‐409.