AMLaP 2004 AixenProvence

1
AMLaP 2004 Aix-en-Provence

• Comprehenders Use Event Structure to Develop
  Discourse-Level Representations
• M.S. Seegmiller, D.J. Townsend, D. DeCangi, K.
  Thomas
• Montclair State University

2
Introduction

• The questions we address
• What is the nature of telicity? I.e. at what
  level in a linguistic representation does it
  appear?
• Specifically, is telicity a binary lexical
  feature or a property that is derived from a
  combination of the semantics of a verb and the
  context in which the verb occurs?
• What role does telicity play in comprehension?

3
Background Assumptions

• Linguistic representations have some
  psychological validity
• Hence the linguistic representations that are
  assigned to words, phrases, and sentences play
  some role (however abstract) in sentence
  comprehension.

4
Brief Overview of Telicity

• Telicity is the semantic property that expresses
  whether or not an event is bounded. Compare
• John slept soundly (unbounded, atelic)
• Mary found her keys (bounded, telic)
• Note that some sentences describe states rather
  than events
• Lucy prefers mango ice cream

5
Self-Paced Reading

• Materials 16 sets of verbs. Each set contained
  one verb in each combination of verb telicity and
  verb subcategorization (pi vs. to). The verbs
  occurred in reduced and unreduced relative
  clauses (RR vs. UR). Example
• The actor (that was) applauded/admired/tripped/cap
  tured on the stage broke the mood of the
  audience.
• Procedure. Participants (N64) read the sentences
  one word at a time with the word centered on the
  screen. Comprehension questions about the content
  of the sentence followed 33 of the trials.
• Data collected self-paced word reading times.

6
Self-Paced Reading Results
7
Corpus Evidence

• We wondered whether a verb's telicity can be
  predicted from its semantics
• Tom fell (telic)
• Tom slept (atelic)
• The question does the semantic telicity
  predicted from a verb's meaning determine the
  telicity of the larger structure in which it
  occurs?
• To answer this question, we analyzed data from
  two large corpora of English.

8
Corpus Data

• Sources
• The British National Corpus 100 million words,
  written and spoken texts, British usage only
• The Collins Cobuild Corpus, 56 million words
  (available on line), written and spoken texts,
  British and American English
• Procedure
• Examination of approximately 40 examples of each
  verb (N64) in context
• Classification of telicity of each sentence

9
Corpus Results

• Of the 2,722 sentences, 2,343 had the same event
  structure classification as the verbs they
  contained. The remaining 379 sentences, or 13.9,
  had a different event structure.
• However, it is usually easy to construct examples
  in which the telicity of a verb is modifiable,
  e.g., He pushed a cart to the store He built a
  house for 20 years but never finished it He
  slept until dawn.

10
Evidence from Speakers' Intuitions

• To test whether native speakers of English shared
  our intuitions about telicity, we administered a
  Forced-Choice Test of sentences containing our
  verbs followed by either  inNP or ''forNP''
• Rationale The in/for test has been used as a
  diagnostic for telicity. Compare
• Ken built the house in a year (telic)
• Ken walked the dog for an hour (atelic)

11
Forced-Choice Test

• Procedure
• Sentences were constructed containing our verbs
  with a prepositional phrase (PP) of the form
  inNP or forNP.
• Sentences were presented in pairs. For example
• John built a house in a year
• John built a house for a year
• Participants (N55) were asked to judge which
  sentence, if either, was better.

12
Proportion of Preferred Sentences on
Forced-Choice In/For Acceptability Test

• in for both
  neither
• Telic .31 .32 .16
  .21
• Atelic .05 .69 .17
  .09

13
Discussion of Forced-Choice Results

• Atelic verbs elicit a strong preference for
  forNP temporal phrases (.69 vs. .05).
• Telic verbs do not elicit such a preference (.31
  vs. .32).
• Conclusion semantically telic verbs are easily
  modifiable, suggesting that telicity is a
  compositional property.

14
Probe-Recognition Test

• The three kinds of evidence discussed so far fail
  to support the view that telicity is an inherent
  lexical property that the parser uses immediately
  to assign structure.
• The next question Do comprehenders integrate
  inherently bounded events more readily into a
  conceptual representation of discourse?
• If so, is it verb telicity or sentence telicity
  that defines boundedness?

15
Probe Recognition Procedures

• Participants (N50) read a story one sentence at
  a time until a probe appeared.
• The participants task was to indicate whether
  the probe is consistent with what they had read
  before.
• We measured the time between the onset of the
  probe and the participants response.

16
Probe Recognition Sample

• The tornado struck the school with little
  warning.
• It was destroyed.
• The firemen didnt think there would be many
  survivors.
• Suddenly, one of the men heard voices in the
  wreckage.
• The firemen rescued a survivor.
• ltT1gt (Target Probe rescue survivor)
• Many of the bystanders helped.
• Some people brought food.
• They were all anxious for the children.
• ltT2gt (Target Probe rescue survivor)
• When the firemen got to the kids, they found them
  safe in the basement of the school.
• (Magliano Schleich, 2001. Thanks to Dr.
  Magliano for providing his materials.)

17
Probe Recognition Procedures

• A probe appeared every 2-4 sentences. The
  critical test points were immediately after the
  target sentence vs. three sentences later.
• Our materials used 16 sentences that differed in
  preference for in- vs. for-temporal phrases.
• The independent variables were verb telicity
  (telic/atelic), object specificity (a
  survivor/survivors), and probe location
  (immediate/3 sentences later).
• Prediction If comprehenders use inherently
  bounded events as anchors around which to
  integrate the content of a story (according to
  the iconicity assumption, cf., Dowty, 1986
  Magliano Schleich, 2000 Zwaan, 1996), response
  times will be faster for telic sentences than for
  atelic sentences.

18
Mean Probe Recognition Times in ms (SE in
parentheses)
19
Probe Recognition Results

• Probe recognition times were faster for sentences
  with telic rather than atelic verbs (1435 vs.
  1569 ms), F1 (1, 49) 13.6, MSe 270728, p lt
  .001, F2 (1, 14) 1.32, MSe 575799, p gt .10.
• There was an interaction between verb telicity,
  object specificity, and probe position, F1 (1,
  49) 4.3, MSe 180388, p lt .05, F2 (1, 14
  3.60, MSe 131777, p lt .10. This interaction
  appears in the table in terms of mean response
  times for correct probe recognition responses.
• Immediately after the target sentence, response
  times for telic verbs were faster with singular
  object NPs than with bare plural object NPs (1242
  vs. 1385 ms).
• Three sentences after the target sentence,
  response times were faster for telic verbs
  regardless of object specificity (1661 and 1446
  ms for telic vs. 1734 and 1747 ms for atelic).

20
Discussion of Probe Recognition Results

• With an immediate probe, both telicity and object
  specificity affect reaction times, with object
  specificity having (if anything) the larger
  effect.
• With a late probe, response times were faster
  with telic than with atelic verbs, regardless of
  object specificity.
• Conclusion verb and sentence telicity affect
  discourse comprehension.

21
Summary

• Data from self-paced reading experiments show
  that subcategorization plays an early role in the
  garden path effect but telicity does not.
• Corpus evidence on modifications of telicity is
  inconclusive on the question of whether telicity
  is an inherent property of verbs.
• Speakers' judgments show that preference for
  inNP vs. forNP phrases does not align with
  semantic telicity of verbs.
• Data from probe recognition experiments indicates
  that telicity plays a role at the level of
  discourse representation.

22
Conclusions

• Telicity does not seem to be a fixed lexical
  property of verbs
• Rather, it is assigned compositionally to the
  predicate, the sentence, or the stretch of
  discourse that contains it
• Thus comprehenders cannot make immediate use of
  telicity information based on the semantics of
  the verb rather, they must delay interpretation
  until more contextual information is available.