An experimental investigation of referentialnonreferential asymmetries in syntactic reconstruction

1
An experimental investigation of
referential/non-referential asymmetries in
syntactic reconstruction

• akira omaki
• anastasia conroy
• jeffrey lidz

Quantitative Investigations of Theoretical
Linguistics 3 June 2-4, 2008
2
what do syntactic judgments reveal about the
grammar?
to help us determine the structure of the
grammar
3
making an acceptability judgment
4
what we do
take a judgment from here
to be reflective of here
5
acceptability judgments
assume we find a construction where a factor, X,
modulates acceptability
we must determine whether X is referenced in the
grammar
or if the variance in acceptability can be
described by the participants ability to create
a context to make a judgment
grammatical hypothesis
context hypothesis
6
string meaning 1 meaning 2
?
string
string
?
grammar
meaning 1 meaning 2
meaning 1 meaning 2
?
context
meaning 1 meaning 2
meaning 1 meaning 2
grammatical hypothesis
context hypothesis
7
acceptability judgments
grammatical hypothesis
context hypothesis
ultimately, we must ask this question of every
factor that influences acceptability judgments
as a way to identify which factors the grammar
references
8
acceptability judgments
grammatical hypothesis
context hypothesis
we are going to investigate this issue with
referentiality and reconstruction a domain
where it has been claimed that referentiality is
referenced by the grammar
9
outline
10
binding principle A

• reflexives in English
• - Mary believes that Bill kicked himself
• - Bill believes that Mary kicked himself

Accessible
Inaccessible
?
reflexives must be bound in a local domain (i.e.,
within the same clause) (Chomsky, 1981)
11
wh-movement adds a new interpretive possibility

• Wh-argument reconstruction
• Tom wondered which picture of himself S Alex
  saw ___

reconstruction interpretation
surface interpretation
Tom
Alex
12
syntactic reconstruction and binding
Tom wondered which picture of himself S Alex
saw ___
reconstruction interpretation low reading
surface interpretation high reading
13
Experimental study on reconstruction and binding

• Leddon 2006 Leddon Lidz 2006
• Argument-fronting vs. predicate fronting in
  adults (and children) in Truth Value Judgment
  Task
• Argument-fronting (both high low )
• Miss Cruella knew which picture of herself Janie
  put up.
• Predicate-fronting (high , low )
• Mr. Whale knew how happy with himself Mr. Walrus
  was.

14
When only one of the interpretations is true in
the context
Both high low readings accepted for argument wh
Only low readings accepted for predicate wh
Leddon 2006
15
referentiality and reconstruction
both interpretations are not always
available it has been argued that
referentiality matters for reconstruction Look
at amount wh-questions (how many x)
(e.g., Heycock, Fox Nissenbaum)
16
referentiality background

• How many people did Jon decide to hire?
• Referential how many is the set of people being
  hired? ? existential presupposition
• Non-referential inquiring about the number to be
  hired

Non-referential
referential
17
referentiality and reconstruction

• LF representations for How many people did Jon
  decide to hire?
• referential ? cardinality of a subset of a
  presupposed set
• How many people1 did Jon decide to hire t1
• Non-referential ? simply inquiring about number
• How many1 did Jon decide to hire t1 people

High reading
low reading
18
referentiality and reconstruction
On the surface they look identical in English,
but these two readings can be expressed in these
word orders in other languages Referential Comb
ien de chansons vas-tu chanter? How many of
songs will you sing? Non-referential Com
bien vas-tu chanter de chansons? How
many will you sing of songs?
19
creation verbs non-referential readings

• creation verb cause x to be in existence
• How many pictures will you draw?
• For what number x, you will cause there to be x
  many pictures in existence? (reconstructed)
• For what number x, there are x many pictures you
  will cause there to be in existence? (surface)
• Surface readings require existential
  presupposition, but this is incompatible with the
  semantics of creation verbs
• only reconstructed (low) reading allowed for
  creation verbs

20
back to reconstruction and binding

• Fox and Nissenbaum (2004) argue that creation
  verbs block binding of reflexive in the surface
  position
• (see also Heycock 1995, Kroch 1989, Longobardi
  1991, Sportiche 2006)
• Non-creation verb, referential
• OKI asked John how many pictures of himself Mary
  is likely to look at t.
• Creation verb, non-referential
• I asked John how many pictures of himself Mary
  is likely to draw t.

21
summary on referentiality

• referential amount Qs asking for a subset of a
  presupposed set ? surface reading
• non-referential amount Qs asking about the
  number ? reconstructed reading
• creation verbs force reconstructed reading due to
  its semantics
• when creation verbs are present, binding
  possibilities are argued to be restricted to low
  reading

22
outline
23
question about the judgment

• Fox and Nissenbaum reconstruction is forced
  because the semantics does not match- surface
  scope (i.e., a grammatical constraint)
• another possibility is that in the judgment, one
  cannot construe the proper semantic context
• therefore, it is a question whether this judgment
  represents something about the grammar

24

• we need to find out

grammatical hypothesis
context hypothesis
claims that the restrictions on interpretation
derive from the grammar (semantics of creation
verbs)
claims that the restrictions on interpretation
derive from the ability to create the relevant
context
25

• .. One needs to know how to tease apart the two
  potential interpretations. The most
  straight-forward way is to consider various
  scenarios for which of the two sentences would
  have different truth values. We think that this
  strategy can be employed however, the strategy
  is fairly involved and we will try to bypass it
  here. Fox and Nissenbaum (2004)

26
referentiality and linguistic judgment
acceptability judgments are quick and easy, but
it is difficult to control the availability of
the context that the participant creates
an experimental scenario allows the researcher to
create the context (so it is available and
consistent across speakers), and then obtain a
judgment
27
outline
28
experiment overview
we want to determine whether the prohibition of
non-reconstructed readings with non-referential
interpretations is a result of the grammar
therefore, we need to test creation/non-creation
verbs (which reflect referential/non-referential
readings) with reflexives (so we can determine
whether reconstruction has occurred) to determine
the range of allowed interpretations
29
experiment overview
we will do this in an experimental scenario that
makes both the referential and non-referential
interpretations equally available
Tom wondered how many drawings of himself Alex
loved to look at
30
experiment overview
our experiment tests the referentiality contrast,
using pairs of sentences, as below the
experimental context provides two possible
antecedents for the reflexives to test
reconstruction
referential, non-creation verb Tom wondered how
many drawings of himself Alex loved to look at
non-referential, creation verb Tom wondered how
many drawings of himself Alex needed to draw
31
the issue

• we want to know if referentiality affects
  reconstruction in the grammar
• or, if the judgment is due to failure to construe
  a possible semantics
• therefore, we need to test in a scenario where
  both interpretations are possible

32
desiderata of experimental design
need to make sure we can tell the difference
between the two readings (reconstructed and
non-reconstructed) cannot ask participants for
explicit judgments about high/low
interpretations, as to avoid meta-linguistic
effects
33
features of experimental design
we will ask the participants to answer the target
question
referential, non-creation verb Tom wondered how
many drawings of himself Alex loved to look
at. Do you know how many?
in a context where the numerical answer directly
reveals interpretation this avoids
meta-linguistic judgments task also alleviates
burden associated with creating context
34
picture gallery task
there are two pictures of Tom in the gallery and
three pictures of Alex this sets up the unique
numerical answer, and the referential
interpretation of the reflexive the art gallery
should be full, so Tom needs three more and Alex
needs two more this sets up the non-referential
interpretation
35
picture gallery task
Tom has work to do, but needs to know how the art
gallery is doing this sets up the need for Tom
to wonder Alex is sent to find out
36
picture gallery task
Alex loves the pictures in the gallery and counts
what is needed this sets up the different
numbers for referential and non-referential Alex
doesnt like the pictures that are
outside contrast set for referential condition
37
target conditions
referential condition Tom wondered how many
drawings of himself Alex loved to look at. Do
you know?
non-referential condition Tom wondered how many
drawings of himself Alex needed to draw. Do you
know?
38
target conditions
referential condition Tom wondered how many
drawings of himself Alex loved to look at. Do
you know? HIGH 2 LOW 3
non-referential condition Tom wondered how many
drawings of himself Alex needed to draw. Do you
know? HIGH 3 LOW 2
39
features of experimental design
the participants answer reveals interpretation
of the reflexive
referential, non-creation verb Tom wondered how
many drawings of himself Alex loved to look
at. Do you know how many?
because the reflexive must be locally bound, this
is a direct indication of reconstruction task
allows us to obtain a judgment in a way that is
easy for the participant
40
predictions
grammatical hypothesis
context hypothesis
referentiality in grammar
referentiality about context
In referential condition, either reading is
permitted In non-referential condition, only low
reading available
either reading permitted in both conditions (no
difference between conditions)
41
ambiguity resolution
because we are looking at ambiguity, we need to
know what interpretations to expect to make
explicit predictions evidence from offline
experiments shows that adults prefer the high
interpretation when available (Leddon 2006)
42
Leddon 2006

• TVJT
• Only using argument (referential) wh-phrases
• Contexts license both high and low readings

43
When both interpretations are true in the context
Adults prefer the high reading over 90 of the
time
Leddon 2006
44
predictions
grammatical hypothesis
context hypothesis
referentiality in grammar
referentiality about context
In referential condition, either reading is
permitted In non-referential condition, only low
reading available
either reading permitted in both conditions (no
difference between conditions)
45
predictions
high readings
grammatical hypothesis
context hypothesis
gt 50
gt 50
referential Non-ref
gt 50
lt 50
46
predictions
high readings
grammatical hypothesis
context hypothesis
referential Non-ref
different
same
47
Experiment design
Version One
Version Two
3 warm up items 2 referential target items 2
non-ref target items 2 control items
3 warm up items 2 referential target items 2
non-ref target items 2 control items
Pseudo-random order
Pseudo-random order
Counterbalanced across participants
48
control conditions
referential control Tom wondered how many
drawings of Alex were in the gallery. Do you
know? 3
non-referential control Tom wondered how many
drawings of Alex were needed at the gallery. Do
you know? 2
49
counterbalanced measures
order of presentation number associated with
high and low readings side characters appeared
on
50
participants

• 21 adults
• native English speakers
• divided evenly between versions

51
predictions
high readings
grammatical hypothesis
context hypothesis
gt 50
gt 50
referential Non-ref
gt 50
lt 50
different
same
52
experiment results
both conditions significantly above chance
(plt.0001) not different from each other (pgt 0.1)
controls 93 correct
53
predictions
high readings
grammatical hypothesis
context hypothesis
gt 50
gt 50
referential Non-ref
gt 50
lt 50
different
same
54
outline
55
discussion

• no difference in high readings between
  conditions, both above chance
• recall the predictions
• grammatical hypothesis only low reading in
  non-referential condition
• context hypothesis ambiguous in both conditions
• therefore, it does not appear that referentiality
  grammatically constrains interpretation

56
alternative interpretation?
non-referential condition Tom wondered how many
drawings of himself Alex needed to draw.

• virtual (if not actual) existence may be
  sufficient for existential presupposition
• (sportiche 2006)

Not in actual existence
57
string meaning 1 meaning 2
?
string
string
?
grammar
meaning 1 meaning 2
meaning 1 meaning 2
?
context
meaning 1 meaning 2
meaning 1 meaning 2
grammatical hypothesis
context hypothesis
58
acceptability judgments
acceptability judgments are not directly
reflective of the grammar!
59
acceptability judgments
grammatical hypothesis
context hypothesis
ultimately, we must ask this question of every
factor that influences acceptability judgments
as a way to identify which factors the grammar
references
60
acknowledgments

• thanks to members of the CNL lab at the
  University of Maryland for their comments and
  suggestions!

61
Experimental study on reconstruction and binding

• Leddon 2006 Leddon Lidz 2006
• Argument-fronting vs. predicate fronting in
  adults (and children) in Truth Value Judgment
  Task
• Argument-fronting (both high low )
• Miss Cruella knew which picture of herself Janie
  put up.
• Predicate-fronting (high , low )
• Mr. Whale knew how happy with himself Mr. Walrus
  was.

62
When only one of the interpretations is true in
the context
Both high low readings accepted for argument wh
Only low readings accepted for predicate wh
Leddon 2006
63
previous research on processing
Omaki et al. (2007) On-line preference for high-
reading (cf. Frazier et al. 1996)