Naturalness of lexical alternatives predicts time course of scalar implicatures Judith Degen Michael K. Tanenhaus Department of Brain and Cognitive Sciences, University of Rochester
Abstract
Introduction
We provide evidence that the speed of computing a scalar implicature (from some to some but not all) depends on the naturalness of some and its lexical alternatives and the speed with which those alternatives become available. We argue that this evidence is inconsistent with the literal-first hypothesis (Huang & Snedeker, 2009).
Default or context-driven inferences? Conflicting evidence (Bott & Noveck, 2004, Huang & Snedeker, 2009, Grodner et al., 2010)
Scalar implicature Alex: Did you submit your paper? Thomas: Some of the sections are written. Not all of the sections are written.
Common problematic assumption: Alternatives to some are invariant over contexts. Hypothesis: Set of alternatives differs across contexts. Speed of implicature depends on 1. naturalness of some and lexical alternatives (given a particular set size)
Experiments 1 & 2
2. speed with which alternative lexical items become available (given a particular set size) We have shown this to be the case for response times (Degen & Tanenhaus, 2011).
Materials and procedure
Experiments: 1. Naturalness of some is low when used with unpartitioned set, where all is an alternative. It is also low in the subitizing range (where number terms are rapidly available) even when number terms are not among the experimental items.
Task: Rate naturalness of statement, given scene (7-point scale or FALSE) IVs: quantifier and set size in lower chamber
2. Adding number terms to the experimental items (i.e. explicitly introducing alternatives) decreases the naturalness of some only in the subitizing range.
Exp n Quantifiers Set sizes 1 120 some, all, none 0 - 13 2 240 some, all, none and 0 - 13 number terms one, two, three, . . ., twelve
3. In eye-tracking: availability effects for small sets and naturalness effects for big sets.
Experiments 3a & 3b Materials and Procedure (n=37)
Results Experiment 1 & 2 ●
●
●
●
●
● ●
6 ● ●
The naturalness of some is
● ●
● ●
• lower for small sets (1-3) than in preferred range (4-7) (β=-0.79, SE=.13, p <.001)
●
●
●
Mean Rating
5
3 Number terms ● ●
absent (Exp 1) present (Exp 2)
Quantifier
2
●
exact some
1 0
1
2
3
4
5
6
7
8
Exp n Task 3a 60 Rate naturalness 3b 37 Click on mentioned gumballs if statement correct, on central button otherwise
• lower than number terms for small sets (1-3) and for unpartitioned set
4
9
10
11
12
Set size
13
IVs (Exp. 3a & 3b) contrast presence in lower chamber target set size (2 or 6) quantifier (some/all/two/six)
• decreased only for small sets (1-3) when adding number terms as explicit alternatives (β=-0.49, SE=.16, p =.001)
DV Rating Judgments & eye movements
Garden-path trials (some with unpartitioned set) included to assess implicature rate.
Results Experiment 3b SMALL SET TARGET
Garden-path trials: 51% central button clicks (reflecting implicature)
6 1.0
quantifier onset
5
Responder types: 43% semantic (03 implicature responses), 54% pragmatic (5-8 implicature responses)
2 1 0
Proportion of fixations
all two/six some
3
COLLAPSED
Quantifier
0.6
0.4 Quantifier all some two
0.2
2
Target set size
optimal breakpoint locations
6 0.0
0
200
400
600
800
1000
1200
1400
1600
0.4 Quantifier all six some
0.2
0.0
0
200
400
600
B REAKPOINT ANALYSIS over window [Q-300ms, Adj+400ms]
1.0
Proportion of fixations
PRAGMATIC
Number terms: Only one breakpoint (BP) for both two (Q+150ms) and six (Q+250ms)
0.6
0.4 Quantifier all some two
0.2
0.0
0
200
400
600
800
1000
1200
1400
1600
Time (ms)
Conclusion
1.0
Proportion of fixations
SEMANTIC
0.8
0.6
0.4 Quantifier all some two
0.2
0.0
0
200
400
600
800
Time (ms)
800
1000
1200
1400
1600
Time (ms)
0.8
Rather than always being delayed relative to processing of literal content, the speed of scalar implicature computation depends on multiple cues, among them the naturalness and speed of availability of some and its lexical alternatives.
0.6
Time (ms)
• some less natural than all/number terms for both sizes (β=-0.81, SE=.16, t=-5.19) • no difference in naturalness for number terms used with big and small set, but all is more natural with the big than with the small set (β=-0.16, SE=.06, t=-2.41)
Time window comparison (baseline vs. [Q+200ms, Adj+200ms]): window effect for all early conditions (implicature drawn before adjective onset)
0.8
1000
1200
1400
1600
Small set target results: large intrusion effects (later BPs, more than one BP) from rapid availability of two for both some and all. More pronounced for pragmatic responders. No delay for some relative to all. Big set target results: early BP for some, reflecting implicature computation, but a second later BP reflects relative unnaturalness (Exp 3a). Pragmatic responders’ behavior more closely reflects naturalness results.
Bott, L., & Noveck, I. A. (2004). Some utterances are underinformative: The onset and time course of scalar inferences. Journal of Memory and Language, 51, 437-457. Degen, J., & Tanenhaus, M. K. (2011). Making inferences: the case of scalar implicature processing. In L. Carlson, C. Hoelscher, & T. Shipley (Eds.), Proceedings of the 33rd annual conference of the cognitive science society (p. 3299 - 3304). Grodner, D., Klein, N. M., Carbary, K. M., & Tanenhaus, M. K. (2010). "some", and possibly all, scalar inferences are not delayed: evidence for immediate pragmatic enrichment. Cognition, 116, 42 - 55. Huang, Y., & Snedeker, J. (2009). On-line interpretation of scalar quantifiers: Insight into the semantics-pragmatics interface. Cognitive Psychology, 58, 376-415.
1.0
0.8
Proportion of fixations
4
1.0
0.6
0.4 Quantifier all six some
0.2
0.0
0
200
400
600
800
1000
1200
1400
1600
Time (ms) 1.0
0.8
Proportion of fixations
Mean rating
meanadjectiveonsets 0.8
BIG SET TARGET
Proportion of fixations
Results Experiment 3a
0.6
0.4 Quantifier all six some
0.2
0.0
0
200
400
600
800
1000
1200
1400
1600
Time (ms)
This work was supported by NIH grant HD 27206 and the Spearmint foundation.
