Transcript Syntactic representation - University of Edinburgh

Syntactic representation
Last week
Lexical access:
Split between syntactic information
(lemma) and phonological (word-form)
information
 Two models:

Levelt et al (modular)
 Dell (interactive)

This week
From words to sentences

Where does syntax fit into the model?
Functional processing (grammatical functions)
 Positional processing (constituent structure)


The nature of syntactic representations:

Syntactic priming
Lexical processing
Lexical concepts
(semantic content)

Lemmas
(syntactic content)

Wordforms/lexemes
(morpho-phonological content)
From words to
sentences
Lexical entries that are retrieved must
be combined into a syntactic structure.
Usually hypothesised to involve two
stages of processing (e.g., Garrett, 1980):
Functional processing
 Positional processing

Where do they fit?
What is the functional
level?
Earliest level associated with linguistic processing.


Input is a conceptual representation
Output is a syntactic representation


perhaps a tree structure
built drawing upon lemma information
Level where:


lemma selection occurs
these are associated with appropriate grammatical functions
(e.g., subject, direct object).
give’<postman, letter, doctor>
> GIVE, POSTMAN, LETTER, DOCTOR
> POSTMANSUBJ; LETTERDOBJ; DOCTORIOBJ
What’s the evidence?
Mostly from speech errors:


Phrase/Word exchange errors:
e.g.
Most cities are true of that.
Writing a mother to my letter.
same grammatical category:



usually non-adjacent (81%):


hence grammatical information relevant.
BUT: could this be meaning-related?
hence unordered.
usually within clause:

hence domain of processing is usually one clause.
Evidence from errors.
Exchanging phrases/words are marked for
grammatical function in landing position:
e.g., She gives them the money
NOT: Her gives they the money
(Int: They give her the money)
 hence, not just misordering of words.
Verbs usually agree with actual subject rather
than intended subject.
e.g.
She gives them the money.
Most cities are true of that.
Unordered
representation
Linear order not specified at F-level.
Not obvious for English:

grammatical roles and order largely the
same:
SUBJECT = first NP in sentence
 DIRECT OBJECT = second NP in sentence
e.g., ISUBJ like BillDOBJ - BillDOBJ likes meSUBJ


but:
ISUBJ can’t stand BenDOBJ, but BillDOBJ ISUBJ like
Unordered
representation (2)
In other (more inflected) languages, grammatical
function and linear order more easily separable:

Junge = boy; Mann = man; Buch = book
Der JungeSUBJ gab dem MannIOBJ das BuchDOBJ
Dem MannIOBJ gab der JungeSUBJ das BuchDOBJ
Das BuchDOBJ gab der JungeSUBJ dem MannIOBJ
Das BuchDOBJ gab dem MannIOBJ der JungeSUBJ
= The boy gave the man the book
The positional level
Second level associated with linguistic
processing.
Level where:
lemmas and associated grammatical
functions are converted into phrase structure.
 I.e, individual words are structured into larger
linguistic units.
 Essentially, syntax as most people think of it

The positional level
GIVE(POSTMANSUBJ; LETTERDOBJ; DOCTORIOBJ)
>
What evidence is there?
Theoretical linguistics:

Substantial theoretical backing for level of
structure defining hierarchical/linear relationships
between abstract categories (phrase structure).
Speech errors:

Sound exchanges:

caught tourses (taught courses),
poppy of my caper (copy of my paper)
Distance constraints: Garrett (1980) found 87% originate
within same phrase.
Pausing/intonation data:

Correlation between pauses and phrase structure.
E.g., Grosjean, Grosjean & Lane (1979)
Syntactic priming
Experimental evidence for abstract syntactic
processing: syntactic priming (aka structural
persistence) effects.
Basic observation:


Speakers repeat syntax in spontaneous speech.
Schenkein (1980):
A: Cor, the noise downstairs, you’ve got to hear
and witness it to realise how bad it is.
B: You have got to experience exactly the same
position as me, mate, to understand how I feel.
More examples
Levelt & Kelter (1982):


A: What time do you close?
B: Six o’clock
A: At what time do you close?
B: At six o’clock
Interpretational problems:


Lexical repetition
Question-answer sequences
Why is this interesting?
Priming effects:
Processing one stimulus is affected by
prior processing of another, related
stimulus.
e.g.,
NURSE - DOCTOR
vs
BREAD – DOCTOR

DOCTOR is faster after NURSE than after
BREAD
The logic of priming
Depends upon processor recognising
relationship between two stimuli:
informative about representation.
Respond faster to DOCTOR after processing
NURSE because using some of same
representations/procedures.
 DOCTOR and NURSE are related (only) in
meaning.
 Therefore, some aspect of lexical processing is
concerned with meaning.

Priming and representation
Priming experiments
Process stimulus with particular characteristics.
 Process subsequent stimulus which is related
along only one dimension.
 If priming effect found (residual activation from
processing first stimulus affects processing of
second), then:



Processor must be sensitive to that dimension of
structure.
So, if production involves stage of abstract
syntactic representation, should find priming
effects based upon syntactic structure.
Experimental
demonstrations
Bock (1986)
Running recognition memory task:
Subjects make recognition decision for
stimuli.
 Ostensibly to aid memory, subjects repeat
sentences and describe pictures.
 Repeated sentences = primes
 Picture descriptions = targets

Bock (1986)
Example trial:

Subject repeats sentence:
The rock star sold some drugs to the undercover agent
[Prepositional Object]


Makes recognition decision:
No
Subject describes picture of girl handing paintbrush to
man standing on stepladder:
The girl is handing a paintbrush to the man [Prep Object]
The girl is handing the man a paintbrush [Double Object]
Bock (1986)
Tendency to repeat structure:




After repeating active sentence, produce active
description;
After repeating passive, produce passive;
After repeating Prepositional Object [PO] (verb
something to someone), produce PO;
After repeating Double Object [DO] (verb
somebody something), produce DO.


The rock star sold the undercover agent some cocaine >
The girl gives the man the brush
No open class (= content) words in common.
Excluding other
explanations
Based on repetition of closed class
words?
e.g. by in passives, to in PO
 Bock (1989):
The secretary baked the cake for the boss
The secretary gave the cake to the boss
> The girl is handing the brush to the man
- Hence, representations not specified for
open class or closed class words
Bock & Loebell (1990)
Based upon event roles?

Primes and targets involving ‘same’ (?) syntax but
different roles:
By-phrase can specify location or agency (who did
action)
The foreigner was loitering by the traffic light
> The boy is being woken by the alarm clock

• Based upon metrical structure?
• Primes and targets involving different syntax but
same metrical structure:
* Susan bought the book to study
Susan brought the book to Stella
> The girl gives the brush to the man
What are positional
representations?
Bock & colleagues:
Abstract - not specified for lexical content;
 Specified for syntactic category.
 ‘Hierarchical configurations of sentences’

Features of positional
representations
Pickering & Branigan (1998):

Sentence completion task
PRIME: The bus driver gave the change…to the passenger
Or: The bus driver gave the passenger…a ticket
TARGET: The surgeon handed…
…the nurse the scalpel (DO)
…the scalpel to the nurse (PO)
…out leaflets (Other)
Features of positional
representations

Manipulations of verb form:

Same vs different tense (hands/handed)

Same vs different number (hands/hand)

Same vs different aspect (hands/is handing)
Results
Priming occurred whether verb varied or not
between prime and target;
But stronger effects if verb remained the
same.
Changing the form of the verb didn’t affect
priming:



Give  gave (tense)
Was giving  gave (aspect)
Gives  give (number)
So the syntactic representations aren’t
specified for verb form.
Lexical repetition effect
• Pickering and Branigan (1998)
 Priming is tied to a syntactic rule
 syntactic rules are linked to verb lemmas
 repeating a verb increases priming
•
Lemma Stratum: encodes syntactic information
NP_NP
NP_PP
combination
Verb
GIVE
Syntactic
category
- GIVE is linked to a PO rule and a DO rule
Features of positional
representations (2)
Theoretical linguistics:

Phrase Structure rules define local trees:
Mother node and daughter node
 E.g. VP -> V NP PP

Features of positional
representations (3)
Positional representations defined in terms of
local trees:

Bigger and smaller structure not specified.
Does priming occur over local trees?

Pickering & Branigan (1998):


PO primes PO, DO primes DO – irrespective of the
structure of NPs.
e.g. The postman gave the letter to the doctor
> The surgeon gave the sharp scalpel to the nervous
nurse
Local trees
Local trees (2)
Branigan, Pickering, McLean & Stewart (2004):

Varied the syntactic context of the prime and target:
The man knew that the postman handed the boy the parcel
> The surgeon handed the nurse the scalpel

The postman handed the boy the parcel
> The consultant suspected that the surgeon handed the nurse the
scalpel


Priming occurred despite changes in context – so the
same processes are involved in producing main and
embedded clauses
Pickering, Branigan &
McLean (2002)
Two possible models of syntax:

S -> NP VP
encodes dominance relations (S is mother of
NP and VP – higher in the tree)
 encodes linear precedence relations (NP
precedes VP in word order)

During production, constituent structure
could be generated in one stage or in
two stages.
Two models
One-stage model:

Phrases are assembled into structure that is
specified for both phrasal composition and linear
order:

S -> NP VP
Two-stage model:

Dominance relations are defined at first stage,
then at subsequent stage placed in linear order:


Dominates(S,NP,VP)
NP < VP
Heavy NP Shift
PO:


The girl gave the wet paintbrush to the man
V NP PP
Shifted:


The girl gave to the man the wet paintbrush
V PP NP
DO:


The girl gave the man the wet paintbrush
V NP NP
PO and Shifted have exactly same constituents,
but in different order.
Method
Spoken and written sentence
completion:
The girl gave the man… (DO)
 The girl gave the paintbrush…(PO)
 The girl gave to the man... (Shifted)

Target:

The teacher gave…
Predictions
If processing is one-stage, shifted and PO
structures do not involve the same
procedures/representations:

they involve different phrase structure rules



VP -> V NP PP
VP -> V PP NP
(PO)
(Shifted)
Therefore, shifted sentences should not prime PO
sentences:

they have nothing in common
Predictions (2)
If it is two-stage, shifted and PO structures involve
some of the same procedures/representations:

they involve the same dominance rules (though different
linear precedence rules)
 dominates(VP,V,NP,PP)
 V < NP < PP (PO)
 V < PP < NP (shifted)

Therefore, shifted sentences should prime PO
sentences:

they have dominance relations in common
Results
Proportion of PO targets/condition:




PO prime
DO prime
Shifted prime
Baseline prime
.70
.46
.59
.62
Subjects produced significantly more POs
after PO prime - but not after Shifted prime

Shifted prime behaved like Baseline
Conclusions
Results are compatible with one-stage
model:
when producing sentence, fully-specified
constituent structure is determined.
 both dominance (vertical relations in tree)
and order (horizontal relations in tree) are
specified.
 No evidence for separate linearisation
process.

Summary
Functional processing



Unordered representations
Assignment of grammatical functions
Influenced by conceptual factors (next week!)
Positional processing:



Ordered representations.
Capture local syntactic relations.
Specified for grammatical category:



not open or closed class lexical content.
Fully specified for hierarchical structure/linear order.
Not subject to conceptual influences.