You are on page 1of 63

Tipping the Scales:

Incremental biases in the processing of degree-modified absolute adjectives

Nicholas Lester Ling 226 Winter 2013

Gradable adjectives
Locate entities along a scale for a given property scale = ordered set of degrees
OSTRICH

d1

d2

d3

d4

d5

HEIGHT

The ostrich is tall Most importantly, GAs take comparative morphology


The ostrich is taller than the penguin
2

Gradable adjectives
Scale structure
GAs can be classified into two major types on the basis of scale structure (along with some corollary properties)
relative gradable adjectives absolute gradable adjectives

Scale structure typology


Relative gradable adjectives
e.g., short and tall unbounded scale structure (no obvious min or max)

short
zone of indifference

tall

only interpretable relative to a contextually determined comparison class (norm) tall gymnast vs. tall basketball player vs. tall building
4

Scale structure typology


Relative gradable adjectives
Special entailment patterns for antonyms (so-called contrary antonymy) ADJpos ADJneg tall entails not short ADJneg ADJpos short entails not tall * ADJpos ADJneg not tall does NOT entail short * ADJneg ADJpos not short does NOT entail tall Hence, the zone of indifference, which allows for cases in which neither adjective is suitable.
5

Scale structure typology


Absolute gradable adjectives
e.g., clean and dirty bounding of the scale on one or both ends
clean dirty

incomplete

complete

empty

full

Scale structure typology


Absolute gradable adjectives
Crucially, the bounding of a scale indicates a logical minimum or maximum value for that property e.g., full is closed because containers have finite capacity Compare with RGAs like tall and short This means that partially or fully closed scales have default context-independent interpretations

Scale structure typology


Absolute gradable adjectives
The two sides of context-independence in AGAs maximum-standard AGAs Logically denote the maximum degree (0% or 100%) of a property e.g., clean minimum-standard AGAs Logically denote any non-zero or non-maximum degree of a property e.g., dirty
8

Scale structure typology


Absolute gradable adjectives
Special entailment patterns for antonyms
ADJmax ADJmin ADJmin ADJmax ADJmax ADJmin ADJmin ADJmax clean entails not dirty dirty entails not clean not clean entails dirty not dirty entails clean

Distinct from relative GAs

Degree modifiers and AGAs


With uniplex entities:
max-selecting DMs These DMs reference degrees that (a) meet a maximum value (b) approach a maximum value absolutely/completely/almost/nearly almost clean/closed/dry ??almost dirty/open/wet

10

Degree modifiers and AGAs


With uniplex entities:
min-selecting DMs These DMs reference degrees at (a) the transition from zero/maximum or (b) any potentially unlimited intensity (scale must not be bounded) slightly/barely/very/extremely slightly dirty/open/wet ??slightly clean/closed/dry
11

Degree modifiers and AGAs


With multiplex entities (sets)
The effect of mostly-adverbs in predicative constructions minimum-standard adjectives The dishes are mostly dirty = Most of the dishes are dirty (to whatever extent) So-called quantity interpretation Also possible: The dishes are all dirty to a significant extent
12

Degree modifiers and AGAs


With multiplex entities (sets)
maximum-standard adjectives The dishes are mostly clean = The dishes are (all) almost clean So-called degree interpretation Also possible: Most of the dishes are clean

13

Experimental results
Frazier et al. (2008)
Demonstrated difference in behavior between max- and min-AGAs in several experiments Speeded acceptability judgment tasks
slightly + max-standard AGA rated less acceptable Completely + max-standard but min-standard rated less acceptable

14

Experimental results
Frazier et al. (2008)
Eyetracker reading paradigm
slightly + max-standard AGA incurred longer first pass reading times

Paraphrase decision task


mostly + max-standard AGA = higher shares of degree interpretation mostly + min-standard AGA = higher shares of quantity interpretation

15

Experimental results
Other pertinent results (RGAs)
Sedivy et al. (1999) inter alia
GAs are processed incrementally
Subjects narrow down the field of potential referents until a unique referent is identified, considering all potential references along the way Plain red rectangle (Eberhard et al., 1995) Degree fixed before the head noun is reached

16

Experimental results
Other pertinent results (RGAs)
Sedivy et al. (1999) inter alia
GAs are contrastive
Information in the visual scene is integrated into the incremental model Pick up the tall one

17

Experiment 1
Proposed questions
Given a contrastive set, do min-selecting DMs facilitate processing of minstandard AGAs over max-standard AGAs (and vice versa)? to what extent are DMs processed incrementally?
Will min-selecting DMs result in more/longer fixations on images exhibiting min-standard properties, regardless of what type of adjective follows?
18

Experiment 1
Procedure
Eyetracker with auditory stimulus Task = select the image which best fits the auditory stimulus RTs, number/length of fixations, and responses collected

19

Experiment 1
Hypotheses
H1 = RTs higher for min-selecting DM + max-standard AGA than for max-selecting DM + min-standard AGA, maxmax, or min-min H2 = Number/length of fixations higher for max-standard images at max-selecting DM (regardless of AGA type) H3 = Number/length of fixations higher for most contrastive min-standard image at min-selecting DM (regardless of AGA type)

20

Experiment 1
Sample Results: Proportional fixations

21

Experiment 1
Design
deg. mod adj. type min-standard selecting very slightly max-standard selecting completely almost

max-standard (e.g., clean)

min-standard (e.g., dirty)

22

DirtyClean

23

DirtyClean
The dish is very {dirty/clean}

24

DirtyClean
The dish is very {dirty/clean}

25

DirtyClean
The dish is very {dirty/clean}

26

Experiment 2
Proposed questions
To what extent is mostly processed incrementally in predicative AGA constructions?
Will mostly result in more/longer fixations on degreebased or quantity-based sets, or both, or neither?

Given competition between degree-based and quantity-based sets, will mostly + min- or maxstandard AGAs result in selection of quantity or degree sets, respectively (as suggested by Frazier et al.)?
27

Experiment 2
Procedure (same as Experiment 1)
Eyetracker with auditory stimulus Task = select the image which best fits the auditory stimulus RTs, number/length of fixations, and responses collected

28

Experiment 2
Hypotheses
H1 = Number/length of fixations higher for degree images at max-standard AGA H2 = Number/length of fixations higher for quantity images at min-standard AGA H3 = RTs higher for contexts involving competing images (i.e., both a degree and quantity interpretation for the target AGA) than for contexts without competing images H4 = Higher selection rates for degree-based images with max-standard AGAs (and vice versa)
29

Experiment 2
Design
Competition set type adj. type max-standard (e.g., clean) Competition = Present quantity degree Competition = Absent quantity degree

min-standard (e.g., dirty)

30

(without competition)

DirtyClean

31

(without competition)

DirtyClean
The dishes are mostly {dirty/clean}

32

(without competition)

DirtyClean
The dishes are mostly {dirty/clean}

33

(without competition)

DirtyClean
The dishes are mostly {dirty/clean}

34

DirtyClean

(with competition)

35

DirtyClean
The dishes are mostly {dirty/clean}

(with competition)

36

DirtyClean
The dishes are mostly {dirty/clean}

(with competition)

37

DirtyClean
The dishes are mostly {dirty/clean}

(with competition)

38

More stimuli
ABSOLUTE (MAX) clean full complete safe straight pure smooth ABSOLUTE (MIN) dirty empty incomplete dangerous bent impure rough/wrinkled

healthy
dry whole intact perfect bald closed

sick
wet cracked damaged flawed hairy open

39

Recap
Despite having received extensive treatment in formal semantic approaches to adjectival meaning, AGAs have only just begun to receive attention from psycholinguists. Given that
Most studies (formal-semantic or otherwise) investigate AGAs using degree-modifying expressions, and Experimental evidence suggests that language processing is incremental ,
40

Recap
A question arises: what are the incremental effects of degree modifiers and AGAs in comprehension? Also, since psycholinguistic research on RGAs has revealed that contrastive-ness plays a big role in adjective comprehension, Another question arises: how absolute are AGAs when presented in competitive environments
41

Future directions
EEG investigation of DMs and AGAs
Can we induce N400 deflections by manipulating (a) AGA and image type? (b) DM and AGA?

Corpus study of AGAs in natural usage


(a) (distinctive) co-ocurrence with DMs?

42

Thank you.

43

Experimental results
Frazier et al. (2008)
Experiment 1: Effect of min-selecting DMs Procedure: Speeded acceptability judgment task Judgment (acceptable/unacceptable) and RT recorded Materials: 16 target constructions I would say that this table is (slightly/a little) {clean/dirty}.
44

Experimental results
Experiment 1
Conditions
With DM slightly/a little Without DM

Min-standard AGA
Max-standard AGA

N=4
N=4

N=4
N=4

45

Experimental results
Experiment 1
Results
Acceptability ratings
Significant main effects of adjective type and presence of DM Significant interaction of adjective type and DM

RT
Significant main effect of adjective type and presence of DM NO INTERACTION

Conclusion
Acceptability but not processing affected by mismatch of DM- and AGA-type.
46

Experimental results
Experiment 1
Open questions
What mechanism is responsible for the unacceptable judgments of slightly + max-standard AGA?
Difficulty integrating the max-standard scale and the scaletransitional meaning of slightly? Violation of expectations provided transitional effects between slightly and the following AGA?

47

Experimental results
Experiment 2: Effect of max-selecting DMs
Procedure
Speeded acceptability judgment task Completely clause (line 1) followed by but clause with a comparison (line 2) Clauses hidden by Xs, revealed by a response key one by one Judgments (acceptable/unacceptable) and reading times were collected

48

Experimental results
Experiment 2
Materials
16 target constructions These juices are absolutely (im)pure, but the mango juice is more (im)pure than the papaya.

Conditions
With DM in line 1 completely Min-standard AGA Max-standard AGA N=4 N=4 With comparative in line 2 N=4 N=4
49

Experimental results
Experiment 2
Results
Significantly more unacceptable ratings for the completely max-AGA but min-AGA condition. No difference in reading times

Conclusions
Completely + max resists qualification (revealing the default standard as the maximum value) Completely + min allows qualification (interpreted like very)
50

Experimental results
Experiment 2
Open questions
Will this effect hold when the subject has (visual) access to the competing degrees of the properties upon encountering the max-selecting DM + AGA phrase?

51

Experimental results
Experiment 3: Processing of DM + AGA
Procedure
Eyetracking during sentence reading Sentences contained combinations of slightly or completely with min-standard or max-standard AGAs 1 in 4 trials followed by comprehension question Reading times and eye movements recorded

52

Experimental results
Experiment 3
Materials
20 target constructions This table is {completely/slightly} {clean/dirty} right now, and the new bus boy is clearly responsible

Conditions
completely Min-standard AGA Max-standard AGA N=5 N=5 slightly N=5 N=5
53

Experimental results
Experiment 3
Results
Significantly longer reading times for first line (counting all passes before the second line was reached) when it contained slightly and a max-standard (e.g., slightly clean)

Conclusions
Slightly + max-standard AGA leads to comprehension problems in reading, but only after the entire clause has been scanned.
54

Experimental results
Experiment 3
Open questions
Why was there no difficulty locally (i.e., at the adjectival phrase itself)? If completely did not create difficulties in comprehension when paired with min-standard AGAs, how might other max-selecting DMs fare?
for example, is ?almost dirty more difficult to process than completely dirty?

More generally, how well do reading times reflect comprehension?


55

Experimental results
Experiment 4: Sentential effects of mostly
Procedure
Paraphrase selection task Choice of alternatives recorded

Materials
12 target sentences The dishes are mostly clean
A) The dishes were {clean/dirty} to a large degree B) Most of the dishes are {clean/dirty}

56

Experimental results
Experiment 4
Conditions
mostly

Min-standard AGA
Max-standard AGA

N=6
N=6

57

Experimental results
Experiment 4
Results
Max-standard adjectives and min-standard adjectives received significantly different proportions of paraphrases Max-standard = degree interpretation Min-standard = quantity interpretation

Conclusions
Scale structure affects sentence-level processing (e.g., it can affect scope of adverbs)
58

Experimental results
Experiment 4
Open questions
How might subjects respond when competing interpretations are available at the time of processing the adjectival phrase? Does overtly specifying the competing interpretations create the potential for biases in the results?

59

Other experiments
Syrett et al. (2006)
Children vs. adults in adjective interpretation tasks
Experiment 1
Children and adults presented with pairs of objects Puppet requests objects Verbal request varied for adjective type (absolute or relative) and pragmatic presuppositions (uniqueness and existence) Puppet may ask for things that arent there (existence violation) or for non-unique things (absolute adjective when both exhibit the min-standard feature e.g., neither cup completely full) Results: Children worse than adults at rejecting requests with violations adults extremely good (94/100 using their scale)
60

Other experiments
Syrett et al. (2006)
Experiment 1
Findings suggest that the absolute forms are not interpreted as contrastive except between the min- and max-standard properties (multiple degrees of min-standard result in rejection of the request). But what if the decision is forced? It is likely that the participant will choose the most contrastive option.

61

Other experiments
Syrett et al. (2006)
Experiment 2
Acceptability judgments The X is Y
Ex. The cup is full

Items varied in terms of the intensity with which they exhibit the property
Cups of varying fullness Discs of varying spottedness

62

Other experiments
Syrett et al. (2006)
Experiment 2
Results: Adults nearly categorically rejected any nonfull cups labeled full and any non-spotted discs labeled spotted Children approximated the adult trend for full but performed identically with spotted

63

You might also like