Transcript GRS LX 700 Language Acquisition and Linguistic Theory

GRS LX 700
Language
Acquisition and
Linguistic Theory
Week 14. Language disorders
and course overview
Dissociations


Language is developmentally dissociable
from other cognitive functions.
Specific Language Impairment


Cover term for situation where language lags behind
in otherwise normal cognitive development
Williams Syndrome

Cognitive impairment, but language developing
relatively normally.
SLI



Language is slower to emerge
Trouble with some—but not all—
inflectional morphology
Some genetic component, seems to
run in families.
SLI and OI



Optional infinitives in normally developing
children has the appearance of a maturational
constraint.
In principle, it might be delayed in some
populations. In a minor way by environmental
factors, but perhaps in a more significant way
due to genetic factors.
Given the association of SLI and apparent
problems with agreement morphology, perhaps
what is happening in SLI is that the OI stage simply
lasts longer.

Rice & Wexler (1996) explore this—in part to see if this can
serve as a testable “marker” to allow for diagnosis of SLI.
Rice & Wexler (1996)



Subjects: 37 5yo SLI, 45 5yo ND, 40 3yo ND
Methods: spontaneous & elicited
Success on -s, -ed, be, do:


Success on -ing, plural -s, in/on


About half of AM, same or worse than LM
As good as AM, LM (which are as good as each other)
Conclusion: It’s not just a delay of inflection. It’s
related to tense, as we might expect of an
extended optional infinitive stage.

Plural -s vs. 3sg -s differ. Can’t be salience.
Rice, Noll, and Grimm (1997)

German: Same finiteness contingency
as with ND kids
SLI
4;0-4;8 +Fin
V2 239
Vfinal 9
-Fin
2
ND
2;1-2;7 +Fin
V2 604
-Fin
11
72
Vfinal 22
37
Wexler, Schütze, Rice (1998)

Testing subject case (ATOM).
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Rice, Wexler,
Redmond (1999)

Judgment task on agreement



(“Language good/not so good”).
A= measure of discrimination,
1=perfect.
BA=bad agreement, DI=dropped
ing.
QuickTime™ and a
TIFF (Uncompressed) decompressor
are needed to see this picture.
QuickTime™ and a
TIFF (Uncompressed) decompressor
are needed to see this picture.
Pragmatics v. syntax

If SLI is a very specific deficit in the
syntactic/computational model, it also provides
an opportunity to test some of the explanations
that have relied on independent development
of a pragmatics component.

Null subjects: Kids assign “very strong topic” too liberally
and drop subjects with finite verbs.


Schaeffer et al. (2001)
Principle B: Kids allow “guises” (coreferential but not
coindexed references), resulting in apparent violations of
Principle B.

Borgeson (2004)
Null subjects (Schaeffer et al.
2001)

Studied subject-verb agreement, subject
case, and null subjects.
SLI
(4yo)
NDLM
(3yo)
NDAM
(4yo)
Bare
stem
33%
39%
Non-nom Null subjects
subjects
18%
16% (with bare stem)
2-7% (with inflected stem)
16%
16-34%
0%
0%
0%
Borgeson (2004)

SLI kids (vocab. Match, 5-6yo ND)

Principle P errors (MB points to her):


ND: 42%
Principle B errors (e.b. points to her):


SLI: 28%
SLI: 20%
ND: 0%
Also explored the effects of bias in the contexts,
and found ND kids were driven by the bias all the
time for Pr.P errors, but the SLI kids were much less
so. Bias played no role for Pr.B. errors.

But cf. van der Lely & Stollwerck (1997). SLI 9-12yo.
Match/mismatch. Appear to take X-self to be any selfdirected action (chance on B tickles B mismatch for B says M
is tickling himself). Chance on (B says) M is tickling him
(mismatch: self). Results on e.b. washes him difficult to
interpret; SLI kids didn’t do very well. Guasti: problem with
quantifiers? As for difference: task?
SLI as Agreement problem?

Clahsen et al. report German SLI kids quite adept
with tense marking, and quite bad at agreement
marking.


Also found in English (Clahsen et al.), Greek (Tsimpli &
Stavrakaki 1999). But not in Italian (Cipriani et al. 1998).
Not expected if SLI=EOI. Not really clear to me
what the source of difference is between Wexler,
Rice, et al.’s results and Clahsen et al.’s results.


RNG suggest difference may arise in data interpretation,
both in contingency analysis and what was counted as what
(participles, V2), or even in subject selection.
Plus, quite possibly SLI>EOI.
Wexler, Schaeffer, Bol (2004)



Agr: /t/ = [-1, +sg/-past], /Ø/ = [+sg], /n/= []
T: /Ø/ = [ ], /te/=[+past]
Schwa is inserted following phonological rules
present
ik werk
jij werk-t
hij/zij/het werk-t
past
1sg
2sg
3sg
wij werk-en 1pl
jullie werk-en 2pl
zij werk-en 3pl
ik werk-te
jij werk-te
hij/zij/het werk-te
1sg
2sg
3sg
wij werk-te-n 1pl
jullie werk-te-n 2pl
zij werk-te-n 3pl
Wexler, Schaeffer, Bol (2004)



Agr: /t/ = [-1, +sg/-past], /Ø/ = [+sg], /n/= []
T: /Ø/ = [ ], /te/=[+past]
Root infinitives are expected whenever Agr is missing and
T is not past: werk-Ø-n.




(Predicts more RIs in English than in Dutch)
Theoretical question about morphology: does [-past]
mean “+present” or “not +past”?
Makes different predictions about what kinds of “errors”
kids will make during OI stage.
Suppose subject is 3sg, and T is missing. T is spelled out as
Ø. Agr is spelled out either as Ø (if -past=+present) or t (if
-past=not +past). Latter will look like present tense—if
past tense was intended, it will look like a tense error.

So: we can decide this theoretical question on the basis of whether kids
in Dutch make bare-stem errors with 3sg subjects.
Wexler, Schaeffer, Bol (2004)






Agr: /t/ = [-1, +sg/-past], /Ø/ = [+sg], /n/= []
T: /Ø/ = [ ], /te/=[+past]
Dutch is V2. We expect that it’s going to turn out
like German, finite verbs in second position,
nonfinite verbs in final position.
Except: Sometimes a form might still look finite,
even if T or Agr is missing.
So: The predictions are much more complicated.
We can also use this to test whether it is T or Agr
(or perhaps both) that is crucial in making the
verb raise to C in V2 languages.
Wexler, Schaeffer, Bol (2004)



Found: SLI kids producing 15% RIs between ages 4 and 8. About
matched (19%) ND kids holding MLU at 3.
Finiteness/V2 contingency very strong. 1% and 2% in the off-diagonal
for ND, 0.2% and 5% in the off-diagonal for SLI. Shows that it’s Agr
responsible for V to C (not T).
Past tense contexts rare: errors in both ND and SLI: -t, -Ø, (a handful,
4-6) and -en (1). Actually a bit ambiguous about which
interpretation of [-past] wins here. But De Jong (1999) reports lots ot t
errors, “not +past” actually wins.
“Not +past”
Predictions
“+present”
+A-T
-A+T
Adult
Context
+A-T
-A+T
Ø
en
Ø
1sg pres
Ø
en
t
en
t
2/3sg pres
Ø
en
en
en
en
pl, pres
en
en
Ø
ten
te
1sg, past
Ø
ten
t
ten
te
2/3sg past
Ø
ten
en
ten
ten
pl, past
en
ten
en
en
en
infinitive
en
en
Williams Syndrome



The other side of the dissociation.
Severe cognitive deficit, probably
traceable to problems with
visuospatial processing.
But language function is
comparatively very good.
Inflection
 Binding theory
 Passives

Zukowski (2001)

Looked in detail at several structures in WS, and
found much that matched ND kids.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Zukowski (2001)

A few things did differentiate WS kids
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.










Acquisition of semantics


Relatively recently, there has been an
increased interest in the acquisition of
semantics.
One popular topic has been investigating
children’s knowledge of sentences like:


The guy didn’t deliver two pizzas.
For adults, this can mean:

There are 2 pizzas that the guy didn’t deliver


(E.g., after delivering the rest of his pizzas, he had 2 left)
It’s not the case that the guy delivered 2 pizzas

(E.g., he delivered 4 pizzas, or 1)
QR

The availability of the two readings is
generally accounted for in terms of
Quantifier Raising (QR):



A quantificational DP moves (covertly) to adjoin to
the IP.
If there are two quantificational DPs, they can do this
adjunction in either order.
The relative structural positions determine their
relative scope unambiguously.


[Not]k [2 pizzas]i [The guy did ti deliver tk ]
[2 pizzas]i [Not]k [The guy did ti deliver tk ]
Musolino’s (2000) OOI

Musolino (2000) found that kids seem
not to reverse scope relations found in
the surface form (the Observation of
Isomorphism).
Every horse didn’t jump over the fence.
 The detective didn’t find some guys
 The smurf didn’t buy every orange.

Musolino & Lidz (2002)




TVJT: Dramatic preference on the part of
kids (4yo) for the surface scope.
Linear order?
Or structure (c-command?)


The detective didn’t find 2 guys.
Tested kids on the same thing in Kannada (where
negation is at the end). Same result: It’s c-command
that matters.
For a while, this was thought to be an
aspect of children’s competence. Their
grammar, e.g., lacked QR.
Gualmini (2003)

By manipulating the context, however,
Gualmini (2003) showed that kids could
access the “inverse scope” interpretation. It
just has to do what would be a
sensible/felicitous thing to say.

Grover orders 4 pizzas from the Troll, who supposed to
deliver them all to Grover. But the Troll drives too fast
and loses 2.



The Troll didn’t deliver some pizzas.
The Troll didn’t lose some pizzas.
90% accept
50% accept
One points out a discrepancy between the expectations
and what actually happened, one doesn’t.
Hulsey, Hacquard, Fox, &
Gualmini (2004)

The Question-Answer Requirement on TVJ tasks:
The test sentence must be understood as an
answer to the “question under discussion.”





Will the Troll deliver all of the pizzas?
Yes (the Troll will deliver them all)
No (the Troll will not deliver them all)
#There are some pizzas delivered by the Troll.
“Isomorphism” isn’t even really a default.

Tested passives (when compared to actives, teases apart
isomorphism and QAR):


Some pizzas were not delivered.
Some pizza were not lost.
94% (adult 100%)
43% (adult 93%)
Kids and QR, Lidz et al. (2003)

(BUCLD 28) More direct test of whether kids
have QR by looking at constructions where
QR is necessary in order to get the right
interpretation.

Quantifier-variable binding


Kermit kissed every dancer before she went on stage
Antecedent-Contained Deletion (ACD)

Miss Red jumped over every frog that Miss Black did.
Quantifier-variable binding

Subject QNP (QR not necessary)


Object QNP (QR required)


Every dancer kissed Kermit before she went
on stage.
Kermit kissed every dancer before she went on
stage.
Kids about 4;6. Acted like adults, yes
where yes was required, no where no
was required.
ACD

VP ellipsis involves interpreting an “empty VP” as
a copy of the “audible VP” (or leaving an
identical VP unpronounced).


John bought a tape and Mary did too.
ACD: the elided VP is inside the audible one.

MR jumped over every frog that MB did.





Audible: jumped over every frog that MB did [VP]
Elided: jumped over … what?
Infinite regress: MR jumped over every frog that MB jumped
over every frog that MB jumped over every from that…
QR solves the problem, though:
[Every frog that MB [jumped over t]]i
MR [jumped over ti].
ACD

MB jumped over every frog that MR did
QR: relative clause reading
 No QR: coordinated reading?




MB jumped over every frog and MB did too.
Kids about 4;5 act basically like adults.
So, they must have QR. Whatever OOI is
about, it isn’t about kids lacking QR from
their grammar.
Musolino & Lidz (2003)

Adults can be made to act like kids with
respect to isomorphism too, in fact.


Try a context where both interpretations
are true…


Cookie Monster didn’t eat two slices of pizza.
E.g.: CM gets 3, but eats 1.
…justifications given seemed to be
isomorphic (75% vs. 7.5% reverse, the rest
unclear).
Entailments

E.g., Minai, Meroni, & Crain (2004):







Can be used as an argument for an innateness
hypothesis vs. constructional analogy.
Kids (mean 4;10), 95% on #2. 90% on #4. 89% on
#5.


John has a black dog > John has a dog
Every boy has a black dog > Every boy has a dog
Nobody has a black dog < Nobody has a dog
Every black dog caught a cicada < Every dog caught a cicada
Nobody caught every black dog > Nobody caught every dog
Nobody could feed every (big) koala bear
Notice that this is a bit early—weren’t Chien &
Wexler (1990) relying on kids’ inability to do
quantifiers?










Language attrition


It is a very common phenomenon that,
having learned an L2 and having
become quite proficient, one will still
“forget” how to use it after a period of
non-use.
While very common, it’s not very
surprising—it’s like calculus. If L2 is a skill
like calculus, we’d expect this.
L1 attrition



Much more surprising is the fact that sometimes
under the influence of a dominant L2, skill in the
L1 seems to go.
Consider the UG/parameter model; a kid’s LAD
faced with PLD, automatically sets the
parameters in his/her head to match those
exhibited by the linguistic input. L1 is effortless,
fast, uniformly successful… biologically driven,
not learning in the normal sense of learning a
skill.
So how could it suffer attrition? What are you
left with?
UG in L2A


We’ve looked at the questions concerning
whether when learning a second language,
one can adapt the “parameter settings” in
the new knowledge to the target settings
(where they differ from the L1 settings), but
this is even more dramatic—it would seem
to actually be altering the L1 settings.
It behooves us to look carefullier at this; do
attrited speakers seem to have changed
parameter settings?
ItalianEnglish

Italian is a “null subject” language that
allows the subject to be dropped in most
cases where in English we’d use a pronoun


(Possible to use a pronoun in Italian, but it conveys
something pragmatic: contrastive focus or change in
topic)
English is a “non-null-subject” language that
does not allow the subject to be dropped
out, pronouns are required (even sometimes
“meaningless” like it or there). Not required
that a pronoun signal a change in topic.
Italian, null subjects




Q: Perchè Maria è uscite?
‘Why did M leave?’
A1: Lei ha deciso di fare una passeggiata.
A2: Ha deciso di fare une passenggiata.
‘She decided to take a walk.’
Monolingual Italian speaker would say A2,
but English-immersed native Italian speaker
will optionally produce (and accept) A1.
(Sorace 2000)
Reverse errors unattested



Q: Perchè Maria è uscite?
‘Why did Maria leave?’
A: *Perchè Ø è venuto a prederla.
‘Because (Gianni) came to pick her up.’
That is, they don’t forget how to use null
subjects so much as they broaden the
contexts in which they can use overt
pronouns.
Postverbal subjects




Q: Chi ha starnutito? ‘Who sneezed?’
A1: Gianni ha starnutito.
A2: Ha starnutito Gianni.
Native speakers would say A2 due to
the narrow focus; attrited speakers will
produce/allow A1 as well.
L1 attrition



It seems that the acceptability of overt
pronouns (in the L1 “attriters”) broadens
compared to their L1, the acceptability of
null pronouns becomes more restricted.
Pronouns in a null subject language are
marked—they are restricted to particular
discourse contexts ([+topic shift], according
to Sorace).
What seems to happen is that the pronouns
revert to the unmarked case ([±topic shift]
like in English).
L1 attrition


Same goes for postverbal subjects—it
is a marked option for languages, and
the L1 seems to be retreating to the
unmarked.
Like with pronouns, it seems to be not
a question of grammaticality but a
question of felicity.
L1 attrition


Certain areas of the L1 grammar are more
susceptible to this kind of attrition then
others.
Sorace notes that the observed cases of
attrition of this sort seem to be the ones
involved with discourse and pragmatics, not
with fundamental grammatical settings. (The
attrited Italian is still a null-subject language,
for example—null subjects are still possible
and used only in places where null subjects
should be allowed).
L1 attrition


So, we’re left with a not-entirelyinconsistent view of the world.
Parameter settings in L1 appear to be
safe, but the discourse-pragmatic
constraints seem to be somehow
susceptible to high exposure to
conflicting constraints in other
languages.










Language mixing
(Spanish-English)



No, yo sí brincaba en el trampoline when I was a senior.
‘No, I did jump on the trampoline when I was a senior.’
La consulta era eight dollars.
‘The office visit was eight dollars.’
Well, I keep starting some. Como por un mes todos los
días escribo y ya dejo.
‘Well, I keep starting some. For about a month I write
everything and then I stop.’
But it isn’t random…

*El viejo man
*The old hombre
*The viejo hombre

*She sees lo.




The old man
El hombre viejo
Certain mixes are not considered to be
possible by fluent bilinguals.
How can we characterize what mixes are
possible vs. impossible?
Prior efforts



Several proposals have been offered to
account for what are good mixes and what
aren’t, but it appears to be a hard problem.
Very famous attempt by Poplack (1980,
1981):
The equivalence constraint. Codes will tend
to be switched at points where the surface
structure of the languages map onto each
other.
The free morpheme constraint. A switch may
occur at any point in the discourse at which
it is possible to make a surface constituent
cut and still retain a free morpheme.
Poplack


Looking at the constraints on codeswitching of this sorts can help us
understand the nature of (at least fluent)
bilingual language representation.
One odd thing about Poplack’s constraints
is that it implies that part of UG is dedicated
to mixing. The Free Morpheme Constraint
and Equivalence Constraint are only
constraints on mixing two grammars. Is UG
built specifically for bilinguals?
Problems for Poplack

Equivalence and Free Morpheme Constraints:
Accounts for *estoy eatiendo, but leaves
unexplained:




The students habian visto la pelicula italien.
*The student had visto la pelicua italien.
*Los estudiantes habian seen the Italian movie.
Motrataroa de nin kirescataroa n Pocajontas
Ref-treat-vsf about this 3s-3os-rescue-vsf in P.
‘It deals with the one who rescues P.’
Problems for Poplack?




*El no wants to go
*He doesn’t quiere ir.
*No nitekititoc
not 1s-work-dur (‘I’m not working’)
Amo estoy trabajando
not be.3s work-dur ‘I’m not working’
Problems for Poplack


*Tú tikoas
tlakemetl
2sg 2s-3Os-buy-fut garment-pl-nsf
(‘You will buy clothes’)
El kikoas
tlakmetl
he 3S-3Os-buy-fut garment-pl-nsf
‘He will buy clothes’
MacSwan 1999


Perhaps the most currently
comprehensive and promising account,
building on recent developments in
syntactic theory.
One of the basic premises is that
language parameters are properties of
lexical items (not of a language-wide
grammar). E.g., verb-movement is due
to a property of the tense morpheme in
French, not shared by the tense
MacSwan 1999



The broad (“minimalist”) approach to
grammar takes language to consist of
two primary components.
Computational system (builds trees),
language invariant.
Lexicon, language particular.
Functional elements of the lexicon
encode the parameters of variation.
MacSwan 1999

MacSwan’s proposal is that there are no constraints
on code mixing over and above constraints found on
monolingual sentences.


(His only constraint which obliquely refers to code mixing is the
one we turn to next, roughly that within a word, the language must
be coherent.)
We can determine what are possible mixes by
looking at the properties of the (functional elements)
of the lexicons of the two mixed languages.
MacSwan 1999


The model of code mixing is then just
like monolingual speech—the only
difference being that the words and
functional elements are not always
drawn from the lexicon belonging to a
single language.
Where requirements conflict between
languages is where mixing will be
prohibited.
Clitics, bound morphemes

Some lexical items in some languages
are clitics, they depend (usually
phonologically) on neighboring words.
Similar to the concept of bound
morpheme.
John’s book.
I shouldn’t go.

Clitics essentially fuse with their host.


Clitics, bound morphemes

Clitics generally cannot be stressed.
*John’S book
 *I couldN’T go.


Clitics generally form an inseparable
unit with their host.
Shouldn’t I go?
 Should I not go?
 *Should I n’t go?

Spanish no





It turns out that Spanish no appears to
be a clitic (despite spelling
conventions).
¿Qué no dijo Juan? ‘What didn’t J say?’
*¿Qué sólo leyó Juan? (‘What did J only
read?’)
*¿Qué meramente leyó Juan?
(‘What did J merely read?’)
*Juan no ha no hecho la tarea.
Nahuatl amo


In Nahuatl, amo ‘not’ does not appear
to be a clitic.
Amo nio amo niktati nowelti.
Not 1s-go not 1s-3Os-see my-sister
‘I’m not going to not see my sister.’
Spanish-Nahuatl mixing



*No nitekititoc
not 1s-work-dur (‘I’m not working’)
Amo estoy trabajando
not be.3s work-dur ‘I’m not working’
Now, we can begin to make sense of
the difference in possible mixes at the
point of negation between Spanish
and Nahuatl.
MacSwan 1999

MacSwan proposes essentially that it is not possible
to code-mix within a (word-like) phonological unit.
Essentially a restriction on what are “pronouncable”
trees.



Idea: phonology operates as a set of ordered rules which are
ordered differently in different languages—you can’t run both
sets of rules at once, hence the result if you tried would be
unpronounceable.
Since Spanish no fuses with the following verb, it
can’t be followed by a Nahuatl verb.
Since Nahuatl amo does not fuse with the following
verb, it is free to be followed by a Spanish verb.
English-Spanish


This also explains Spanish-English (well,
Spanish-anything)
*El no wants to go

What about English-Spanish?
*He doesn’t quiere ir.

*He doesn’t wants to go.

Agreement





In languages that code agreement between
subject and verb, it also appears that mixing is only
possible where the agreement relationship is not
disrupted.
*He doesn’t quiere ir.
English negation: agreement appears on do.
Spanish negation: agreement appears on the
verb.
You can’t have extra agreement: one subject, one
agreement. They need to match.
Agreement



*Yo nikoas tlakemetl
I 1s-3Os-buy-fut garment-pl-nsf
(‘I will buy clothes’)
*Tú tikoas tlakemetl
you 2s-3Os-buy-fut garment-pl-nsf
(‘You will buy clothes’)
Él/Ella kikoas
tlakemetl
He/She 3s-3Os-buy-fut garment-pl-nsf
‘He/She will buy clothes’
Agreement





Ni-k-koa-s ‘I will buy’
Ti-k-koa-s ‘You will buy’
Ø-k(i)-koa-s ‘He/she wlll buy’
Also relevant: Spanish marks and agrees with gender
but Nahuatl does not distinguish masculine from
feminine.
Spanish pronouns have gender specification. The
Nahuatl verb does not. They can only be compatible
(match) if there is no Nahuatl agreement
morpheme.
Spanish-Catalan-Greek



Spanish and Catalan both have two
genders, masculine and feminine.
Greek has three genders, masculine,
feminine, neuter.
Predicts: Mixing subjects and verbs
between the three languages is only
possible between the gendercompatible languages.
Spanish-Catalan-Greek





Yo vull mengar el dinar (S-C)
Jo queiro comer la cena (C-S)
*Ego vull mengar el dinar (G-C)
*Ego queiro comer la cena (G-S)
…
Mixing and L2A?


Code mixing as discussed so far is generally a
property of the speech of fluent bilinguals (often
native bilinguals) and reflects properties of
universal language knowledge.
We can now return to our old question and ask:
Does the knowledge of second language
learners also have the restrictions on code
mixing? To the extent that this is “part of UG”, is
this aspect of UG active for L2’ers? For the
future—I’m not aware of studies on L2A.









