The effect of context on the processing of typeshifting verbs

1
The effect of context on the processing of
type-shifting verbs

• Roberto G. de Almeida
• Max Knobbout
• Marlies Paenen

2
Overview

• Introduction
• Aim of the paper
• Hypotheses
• Hypothesis 1
• Hypothesis 2
• Hypothesis 3
• Experiments
• Experiment 1
• Experiment 2
• Conclusion
• Discussion

3
Introduction Two views

• Atomic view
• Lexical items are mapped onto atomic, indivisible
  concepts. They are denotational, abstract and
  metalinguistic (see Fodor, 1998).
• Decompositional view
• Lexical items (in particular nouns and verbs) are
  represented in terms of decompositional features
  or templates (see previous week Pustejovsky,
  1995).

4
Introduction Two views

• Consider the sentence Mary killed the dog.
• Atomic view
• MARY, KILL and DIE are abstract concepts.
• Meaning can be obtained by meaning postulates. In
  this case if x kills y, y dies

5
Introduction Two views

• Consider the sentence Mary killed the dog.
• Decompositional view
• Semantic representation
• Maryx CAUSE the dogy DIE
• s(kill) x CAUSE y DIE
• Or in other words, possession of a conceptual
  template implies the possession of its
  constitutive concepts.

6
Introduction Two views

• Atomic view Classical (Fregean) composition.
• This states that the meaning of a whole
  expression is a function of the meaning of its
  atomic constituent parts.
• Decompositional view Enriched composition.
• This states that the meaning of a whole
  expression is determined by its decomposed parts,
  driven by coercion processes.

7
Introduction Two views

• Evidence supporting the decompositional view
• McElree, 2001, who found longer reading times
  for type-shifted verbs.
• Evidence disputing the decompositional view
• Fodor Lepore, 1998, who argue that there is
  no distinction between grammatical knowledge and
  world knowledge. Also, what are the telic
  properties of a rock (a rock to break a
  window)?

8
Aim of the paper

• Find (convincing) empirical evidence which
  supports either the classical view of
  compositionality, or the view of enriched
  compositionality.

9
Hypothesis 1

• Context cancels type shifting operations. The
  interpretation of the VNP construction (e.g.
  started the book) is not a function of the
  nouns telic roles (reading, writing, etc.),
  but a function of the context that decides the
  interpretation of the NP.

10
Hypothesis 1 Example

• Example
• My goat eats anything.
• He really enjoyed your book.

11
Hypothesis 1 Example

• Example
• My goat eats anything context.
• He really enjoyed your book.

12
Hypothesis 1 Example

• Example
• My goat eats anything context.
• He really enjoyed eating your book.
• Note No type shifting took place, it was
  canceled by the context.

13
Hypothesis 1

• My goat eats anything. He really enjoyed your
  book.
• The goat enjoyed eating your book.
• Both these sentences (preferred and type-shifted)
  should have similar (post-verbal) reading times!

14
Hypothesis 2

• Type coercion is a necessary process, and context
  constrains the number of telic roles that may be
  selected for the object NP.

15
Hypothesis 2 Example

• Example
• My goat eats anything. He really enjoyed your
  book.
• Then
• Qualia structure for book becomes
• Telic reading, writing, eating, ....

16
Hypothesis 2

• Type shifted My goat eats anything. He really
  enjoyed your book in his garden.
• Non-preferred The goat enjoyed reading your
  book in his garden.
• Preferred The goat enjoyed eating your book in
  his garden.
• Same colored parts should have similar reading
  times!

17
Hypothesis 3

• Context removes the underspecification of the
  main type-shifting verb (such as began,
  enjoyed), leveling them out with non-preferred
  verbs.

18
Hypothesis 3 Example

• The goat enjoyed eating your book
• The red crosses denote the
• semantic fitting of the concepts.

A B C D E F ?

19
Hypothesis 3

• This semantic fitting can be sped up providing a
  context, leveling them with non-prefferred verbs.
  Thus non-preferred and type-shifted verbs should
  have similar reading times.

20
Experiment 1 (1/4)

• Designed to replicate the effect obtained by
  McElree et al. (2001)
• Similar type-shifting effect expected because
  similar (but slightly modified) materials.

21
Experiment 1 - Method (2/4)

• Participants 69 native English speakers
• Materials and design
• 21 sets of triads (preferred, non-preferred,
  type-shifted)
• Examples of McElree et al. (2001) modified
• The teenager started/read/wrote the novel
• The student started/read/wrote the paper
• Procedure --- ---------

22
Experiment 1 Results (3/4)

• V1 slower reading times type-shift construction,
  but not significant
• V2 no significant differences
• V3 type-shifting constructions significant
  faster than preferred constructions

23
Experiment 1 Discussion (4/4)

• Type-shifting effect obtained by McElree et al.
  is not replicated.
• Reading times are slower overall.
• Lack of difference between preferred,
  non-preferred, and type-shifting constructions.
• This results (together with Taxler et al.s)
  suggest that type-shifting verns are no more
  complex than other verb types.

24
Experiment 2 (1/4)

• Same contrast between three types of
  constructions as in Exp. 1, but preceded by
  contextual information that constraints the
  nature of the event performed by the agent over
  the referring complement NP.

25
Experiment 2 Method (2/4)

• Participants 36 native English speakers
• Materials and design
• 21 sets of triads, the same as in Exp. 1,
  preceded by a paragraph context
• The secretary would always be sure to work ahead
  of schedule. She was asked to work on a memo.
• The secretary began/read/typed the memo.
• Procedure --- ---------

26
Experiment 2 Results (3/4)

• Overall faster reading times for preferred
  constructions
• V1 significant difference between
  type-shifting/non-preferred and preferred
• V2 differences fail to reach significance
• V3 no significant differences

27
Experiment 2 - Discussion(4/4)

• Type-shifting constructions behave similarly to
  non-preferred constructions.
• Both yield slower reading times than preferred
  constructions.
• Contexts seem to set apart preferred and other
  types of verbs.

28
Conclusion (1/4)

• Hypothesis context would act by canceling the
  type-shifting operation.
• The results suggest that there are no
  type-shifting operations whatsoever, if these are
  construed as operations over lexical-semantic
  information.
• Failure to replicate McElree et al.s results and
  lack of difference between non-preferred and
  type-shifting constructions cast doubt on the
  hypothesis that any type-shifting function occurs.

29
Conclusion (2/4)

• Hypothesis context would act by constraining the
  range of telic roles to be selected within the
  lexical entry for the complement noun.
• If there was reduced semantic activity there
  would be a delay in semantic processing of
  type-shifting constructions at positions
  immediately after the verb.
• This is not corroborated by the results. Is
  semantic interpolation involved in the process of
  interpreting type-shifting constructions?

30
Conclusion (3/4)

• Hypothesis context would remove the
  underspecification of verbs, setting apart verbs
  that were preferred for given contexts.
• Assumption that all processes of integration of a
  semantic representation are pragmatic and
  inferential.
• This can account for the pattern of results, and
  can explain McElree et al.s results

31
Conclusion (4/4)

• The experiments presented here show no evidence
  for enriched composition, but can be taken to
  support a classical view of semantic
  compositionality.
• What is taken to be a type-shifting operation
  over semantic structures is a set of inferential
  processes beyond the logical form of an
  expression, together with the denotation of its
  constituent items.

32
Discussion

• Findings are supported by
• de Almeida, Dwivedi, 2009, who revisited this
  paper and propose a redefinition of the nature of
  type-shifting operations. Their main objection to
  McElree is the unclear distinction between
  pragmatic and lexical information.

33
Discussion

• Findings are disputed by
• Many. Next presentation discusses one of them,
  namely McElree, 2004.