Parafoveal processing of the second constituent of compound words in Finnish

Preview:

DESCRIPTION

1. 2. Parafoveal processing of the second constituent of compound words in Finnish. Raymond Bertram 1 , Sarah White 2 , Jukka Hyönä 1. AMLaP 2006 , 31.8.2006, 09:20 - 09:40 Oral session 4: Morphology. Compound words. LASTEN/TARHAN/OPETTAJA/KOULUTUS ‘ kindergarten teacher schooling’ - PowerPoint PPT Presentation

Citation preview

Parafoveal processing of the second constituent of

compound words in Finnish

Raymond Bertram1, Sarah White2, Jukka Hyönä1

AMLaP 2006, 31.8.2006, 09:20 - 09:40 Oral session 4: Morphology

1 2

Compound words

LASTEN/TARHAN/OPETTAJA/KOULUTUS

‘kindergarten teacher schooling’

VANILJA/KASTIKE

‘vanilla sauce’

SIVU/OVI

‘side-door’

Perceptual span

attention

------->

periphery

parafovea: poor acuity

40 40

fovea:goodacuity

20

Effective visual field in reading

Parafoveal processing: 2 critical issues

1. Do we extract semantic information from the parafovea?

Heated debate going on about this issue, most evidence speaks against it (see Rayner et al., 2003, for a review)

2. Are parafoveal words processed in parallel with fixated words?

Serial models (e.g. EZ-Reader, Reichle et al., 2003)Processing Word N before Word N+1, although Word N+1 can

be partly processed while on Word N (in later phase) Parallel models (e.g., SWIFT, Engbert, Longtin, Kliegl, 2002)

Processing Word N + Word N+1 simultaneously

Word N Word N+1

Bertram & Hyönä, JML, 2003, found that access of long compound starts off with access of 1st constituent, purely because of visual acuity reasons

1

vanilja/kastike

Effective visual field & compound words

’vanilla sauce’

Hyönä, Bertram, Pollatsek (MC, 2004) found that, nevertheless, orthographic information is extracted from 2nd constituent while fixating the 1st constituent

Current study

1. Do we extract semantic information from the parafovea in

compounds like vaniljakastike?

=> would imply that extracting information from parafovea is not only an acuity issue, but also linguistically determined

2. Are parafoveal constituents/lexemes processed in parallel with fixated constituents/lexemes?

=> across words more evidence for serial processing => does this extend to within-word processes?

vanilja seokliikastike

kastike

Identical condition: 2nd constituent same throughout

Change condition: 2nd constituent changes after saccade over invisible boundary

1 2

vanilja

1 2

Current study: boundary XP, 4 conditions

Current study: boundary XP, 4 conditions

1. Identical: vanilja/kastike ’vanilla sauce’ 2. SemRelated: vanilja/sinappi ’vanilla mustard’3. SemUnrelated: vanilja/rovasti ’vanilla priest’4. Nonword: vanilja/seoklii ’vanilla nonword’

1. Do we extract semantic information from the parafovea in compounds like vaniljakastike? 2 < 3, 4

2. Are parafoveal constituents/lexemes processed in parallel with fixated constituents/lexemes? => Parafoveal-on-foveal effects

1 2 3 4

5 6

vanilja/sinappiFirst Fixation Duration = Subgaze1 = + : gaze duration before

boundary change

Subgaze2 = + + + : gaze duration after boundary change

Early measure

Late measure

1

1 2

3 4 5 6

First fixation duration on 2nd constituent = 3

Time course of long compound processing

2. Are parafoveal constituents/lexemes processed in parallel with fixated constituents/lexemes? => Parafoveal-on- foveal effects

vanilja/kastike

Method

1. Identical: vanilja/kastike ’vanilla sauce’ 2. SemRelated: vanilja/sinappi ’vanilla mustard’3. SemUnrelated: vanilja/rovasti ’vanilla priest’4. Nonword: vanilja/seoklii ’vanilla nonword’

Semantically related: Lauran mielestä vaniljasinappi kuuluu ...

Lauran mielestä vaniljakastike kuuluu ...

To Laura’s mind vanilla sauce belongs ...

Earliest measure: First Fixation Duration

200

205

210

215

220

225

Identical:kastikesauce

SemRel:sinappimustard

SemUnr:rovastipriest

Nonword:seoklii

Fir

st f

ixat

ion

Du

rati

on

No difference between 4 conditions (F1,2 < 1).

vanilja/kastike3 4

5 61 2

1

Early measure: SubGaze1

270

290

310

330

350

370

Identical:kastikesauce

SemRel:sinappimustard

SemUnr:rovastipriest

Nonword:seoklii

Su

bG

aze1

No difference between 4 conditions (ps > .15).

vanilja/kastike3 4

5 61 2

1 2

Visual-Orthographic

Effect

Early late measure: First fixation duration on 2nd constituent

190

200

210

220

230

240

250

Identical:kastikesauce

SemRel:sinappimustard

SemUnr:rovastipriest

Nonword:seoklii

FF

D o

n c

on

stit

uen

t 2

Main effect, p1,2 < .001 Identical vs other 3 conditions (all ps < .001)

vanilja/kastike3 4

5 61 2

3

Visual-Orthographic

Effect

Lexical-Semantic

Effect

Late measure: SubGaze2

Main effect, p1,2 < .001 Identical vs other 3 conditions, all ps < .001 Other contrasts: 2-4 ps < .001; 2-3 and 3-4, ps < .05

250

290

330

370

410

450

Identical:kastikesauce

SemRel:sinappimustard

SemUnr:rovastipriest

Nonword:seoklii

Su

bG

aze2

4

3

2

1

vanilja/kastike3 4

5 61 2

3 4 5 6

Conclusions

Parafoveal processing: 2 critical issues (1)

1.Do we extract semantic information from parafovea?

If parafoveal area is 2nd constituent of a compound, the answer is yes!

=> While fixating on 1st constituent 1(vanilja), semantic information of 2nd constituent is extracted, leading to faster processing in the late stages of compound processing (after crossing constituent boundary)

Parafoveal processing: 2 critical issues (2)

2. Are parafoveal words/lexemes processed in parallel with fixated words

No! All effects were found in late measures, nothing on firstconstituent. In other words, there were no parafoveal-on-foveal

effects

=> Initial processing pertains to processing of 1st constituent.

Time course of long compound processing

=> preview of 2nd constituent; information of 2nd constituent extracted, from orthographic to semantic

vanilja/kastike1

=> Access 1st constituent: vaniljavanilja/kastike1

vanilja/kastike2 => access of 2nd constituent; ortho-

graphic preview benefits cashed in

vanilja/kastike34 5 => Semantic preview benefit

cashed in; meaning integration of constituents

Implications for eye movement control models

• Semantic preview benefit within compound words contrasts with parafoveal processing across words=> extracting parafoveal information is not only question of visual acuity => linguistic relationship between lexical units important as well!

• Lack of parafoveal-on-foveal effects in line with models of serial processing, e.g. EZ-Reader. => Attention shifts to next word/lexeme after currently fixated word/lexeme has been accessed

Kiitos!

Boundary experiment with 4 conditions

1. Identical: vanilja/kastike ’vanilla sauce’ 2. Semantically related: vanilja/sinappi ’vanilla mustard’3. Semantically unrelated: vanilja/rovasti ’vanilla priest’4. Pronounceable nonword:vanilja/seoklii ’vanilla nonword’

• 1,2,3,4 matched on 1st/2nd constituent length (average 7.5/5.4)• 2,3,4 on visual-orthographic overlap with 1;• 1,2,3 on 2nd constituent frequency (around 200 per million);• 1 vs 2 and 2 vs 3 pretested on semantic relatedness (scale 1-7)

=> 1 vs 2: 5.8 2 vs 3: 1.4

• Semantically related: Lauran mielestä vanilja/sinappi kuuluu ...

• Semantically related: Lauran mielestä vanilja/kastike kuuluu ...

• In order to minimise the possibility of participants consciously noticing display change => boundary located prior to the ultimate letter of the first constituent. • Participants were only included in the analyses if they reported to have noticed no more than five changes

Method

• Latin square design • 28 sentences were filler items without display change• 4 lists of 84 sentences were constructed (56 targets + 28 fillers)• 50% items without display change (14 identical + 28 fillers)• 7 participants were randomly allocated to each list• Participants asked to read for comprehension• Comprehension question after 18 of the 84 sentences• Eye movements monitored by EyeLink 2

Method

Size of parafoveal preview benefit (Identical – Preview Type)

Visually Visually First 2-3 First 2-3First 2-3Dissimilar Similar Identical Identical Identical

Study All Xs Letters Letters +Rest Xs Rest Vis. Rest Vis. Dissimilar Similar

Balota et al. 1985 +31 +8Rayner et al. 1986 +40 +5Lima, 1987, XP1 +18 +1Lima, 1987 XP2 +29 +30Inhoff, 1989a, XP1 +52 +26Inhoff, 1989a, XP2 +53 +38Inhoff, 1989b, XP1 +54 +38Inhoff, 1989b, XP2 +32 +24Inhoff, 1989b, XP3 +22 +16Henderson &Ferreira, 1990, XP1 +5 -6Henderson &Ferreira, 1990, XP2 +11 +1Pollatsek et al., 1992 +43 +25 +10 -17Briihl & Inhoff, 1995, XP1 +38Briihl & Inhoff, 1995, XP2 +41 +30Kennison & Clifton, 1995 +28Inhoff et al., 2000 +91Altarriba et al. (2001) +33 +15

Mean +42 +41 +16 +28 +14 -4

•The effect is much larger than any preview benefit effect before

Further conclusions

• => attention spreads more to the parafovea within a compound word than across two subsequent words!

vanilja/kaefhlavanilja/kastike

101 ms

Hyönä et al, 2004

new door new tune new sorpnew song new song new song 44ms 40 ms 5 ms

Rayner, Balota, Pollatsek, 1986

Change:NoChange:

Implications for processing models

At least for one type of words, we can say: Yes =>before whole-word units, c1 before c2

• Morphological processing modelsAre morphemic units used in the course of processing? If so, how/when?

Whole word access

Morphemic access

VANILJA_KASTIKE

C1 C2

IntC1C2

Method Hyönä et al.

1) Boundary that determined display change was always the constituent boundary.

2) The first two letters of the 2nd constituent were preserved in the change condition and all the other letters were changed to visually similar letters).

3) This was done so that display change was not noticed and to create a visual-orthographic condition that was quite similar to the no change condition

Studying processing of compound words in context using online measures of eye movement behavior

More detailed insight in the role of morphology during complex word processing => Morphological processing models

Specification of eye movement behavior as a function of morphological structure => General eye movement models of reading

Compound words & processing models

Compound words & processing models

• Morphological processing modelsAre morphemic units used in the course of processing? If so, how/when?

Whole word access

SIVU_OVI Morphemic access

VANILJA_KASTIKE

brilliant

• We do extract information from word N+1, while we are fixating on word N => Parafoveal preview benefits

• The information we extract is low-level information (pertaining to word length and orthographic/phonological level)

• We process words in a serial manner: first word N, than word N+1 => We do not find effects on word N as a function of manipulations of word N+1 => No parafoveal-on-foveal effects (cf EZ-READER, Reichle et al., 2003)

Parafoveal processing across words

WORD N WORD N+1

tunesongsorpsongsong Rayner, Balota, Pollatsek, 1986

Compound words & processing models

General eye movement models of reading• How do foveal and parafoveal processes interact?

John Smith is a great groundskeeper.

b. Hyönä et al. (2004) also tested whether readers extract orthographic information from the latter part of the 2nd consti-tuent, at the same time as they process the 1st constituent.

No change condition: v a n i l j a k a s t i k e

Change condition: v a n i l j a k a e f l h a

Effective visual field & compound words

1 2 3 4

If you extract orthographic information from kastike at the same time as you process the first constituent, the change manipulation should affect processing behavior on early measures already => parafoveal-on-foveal effects

Gaze Duration whole word= + + + + +

1

vanilja/kaeflha2 3 4

5 6

vanilja/kastike

First Fixation Duration = Subgaze1 = + : gaze duration before

boundary changeSubgaze2 = + + + : gaze duration

after boundary change

Early measure

Late measure

1

1 2

3 4 5 6

1 2 3 4 65Global measure

Major Findings Hyönä et al.

Change effect, but in later measures only

MeasureNoChange (kastike)

Change (kaeflha)

Change Effect

1st Fixation duration 225 227 2

Subgaze1 260 268 8

Subgaze2 345 426 81 **

Gaze Duration 538 639 101 **

1

vanilja/kaeflha2 3 4

5 6

vanilja/kastike

1

1 2

3 4 5 6

3 4 5 61 2

Orthographic information of latter part of second constituent is picked up during first fixation(s).

However, the fact that the change effects are late suggests that processing of the two constituents is serial (in line with e.g. EZ-Reader)

First constituent frequency manipulation in this experiment yielded solid effects from the first fixation onwards

Conclusions

• With more attention spreading to parafovea, semantic parafoveal processing may take place within compounds

• Visual-orthographic manipulations in Hyönä et al. were subtle => parallel processing may take place (e.g., processing 1st constituent and first letters of 2nd constituent at the same time), but stronger visual-orthographic manipulations are called for

than in case of k a s t í k e vs. k a e f l h a

Current study: boundary XP, 4 conditions

1. Identical: vanilja/kastike ’vanilla sauce’ 2. SemRelated: vanilja/sinappi ’vanilla mustard’3. SemUnrelated: vanilja/rovasti ’vanilla priest’4. Nonword: vanilja/seoklii ’vanilla nonword’

Current study: boundary XP, 4 conditions

1. Identical: vanilja/kastike ’vanilla sauce’ 2. SemRelated: vanilja/sinappi ’vanilla mustard’3. SemUnrelated: vanilja/rovasti ’vanilla priest’4. Nonword: vanilja/seoklii ’vanilla nonword’

1. Do we extract semantic information from the parafovea in compounds like vaniljakastike? 2 < 3, 4

1b. Do we extract orthographic information from the parafovea in compounds like vaniljakastike? 1 < 2, 3, 4

• LASTEN_TARHAN_OPETTAJA_KOULUTUS

• Bertram & Hyönä, JML, 2003

• => Visual Acuity Hypothesis: access of long compound starts off with access of 1st constituent due to visual acuity benefit of 1st constituent over the latter part of the word.

Effective visual field & compound words

1 1

sivu/ovi

2

vanilja/kastike

Recommended