Quantifying the temporal cost of processing garden-path sentences(Summary of Awarded Presentation at the 26th Annual Meeting)

(1)

The Japanese Psychonomic Society

NII-Electronic Library Service

The JapanesePsychonomic Society

The _Jlapanese

]ournat

of

PsychonornicScience

2008.VoL 27,No.1,115-1l6

Summaryof

Awarded

PresentationIP54

Quantifying

the

temporal

cost

of

_processing

garden-path

sentences

Miki

UETsuKii･*,

Kazushi

MARuyA2'

**, _and

_Takao

_SAToi

UnivensiC),

of

Tohyo]

and

The

Jifeei

U>iiversity2

lt

is

well established that English

_garden-path

sentences require more time to be read as

compared tonormal sentences {garden-path phenomenon), In contrast, no clear evidence

for

this

phenomenon isreported

in

thecase of

Japanese

_garden-path sentences, Inthisstudy, we examined

the reading _performance while varying the presentation speed, and estimated the

temporal

cost of

Japanese

garden-path sentences.

The

results showcd that

Japanese

garden-path sentences require

longer

processing time than non-garden-path sentences.

The

estimated cest

is

about 50-1OO ms/

phrase.

Thu$,

the examination with

the

experimenter-paced method revealed the garden-path

phenomenon

in

Japanese.

Key

words: _garden-path,

Japanese,

temporal cost, SAT

function

Garden-path sentences require reanalys{s

due

to the temporal ambiguity of theirsyntactic

construc-tion

{Frazier

& Fodor, 1978).i The reading times

for

garden-path sentences are _generally longer than

those for non-garden-path sentences

(this

is

called

the garden-path phenomenon). However, with

sim-ple

Japanese

garden-path stimuli, the _garden-path

phenomenon is often not

detected

because it is weaker inthe

Japanese

language than inEnglish. No

previous study

has

quantified the temporal cost of

processing garden-path sentences

in

Japanese,

and it

isunclear how heavy the _processing

load

of such

sentences may

be,

Inthis _paper, we measured the

accuracy of sentence _processing

for

various

time

durations and then _quantified the temporal cost of

processing

Japanese

garden-path sentences.

* _Center _for _Evolutionary

_Cognitive

_Sciences,

The

University

of

Tokyo,

3-8-1

Komaba,

guro-ku,

Tokyo

153-8902,

Japan

** _Kazushi _Maruya

_is

_supported

_by

_JSPS.

i

_When

we read the garden-path sentence

L`Haruyo-ga

shuuii-o hajimeta Harumi-ni

tegami-o okutta;d we interpret itas "Haruyo

started calligraphy" till the third phrase.

However, at the

fourth

phrase,

it

is

revealed

that this interpretation iswrong and that the

correct interpreation is "Harumi

started

calligraphy."

Method

Five undergraduate students and postdoctoral

fel-lows

participated

in

this experiment. The stirnuli were 120 _garden-path sentences and

120

non-garden-path sentences, such as those given inexamples

(1)

and

(2)

below. The gardenpath sentences require

syntactic reanalysis at the fourth phrase.

Non-garden-path sentences had the same syntactic con-structions up tothe third_phrase and did not require reanalysis of syntactic stru ¢ture, Both types of

sen-tences consisted of six _phrases,and all _phrases

con-sisted of threecharacters and

four

morae.

(1)

Garden-pathsentence:

Haruyo-ga shuuji-o hajimeta Harumi-ni tegami-o

okutta. "Haruyo

sent a

letter

to

Harumi

who had

started calligraphy."

(2)

Non-garden-path sentence]

Haruyo-ga

shuuji-o

hajimeta

wadai-ni

ga hazunda, "Harumi

was _pleased with the topic

that

Haruyo

had

started calligraphy."

Sentences

were visually _presented _phrase

by

phrase.

Six

differentdurat!ons were examined: 16.7,

33.3,

66.7,100, 150,and 250 mslphrasc. After each

stimu]us sentence, a_question about the stimulus was

presented,

Participants

had

toanswer the _questions

using the two-alternative forced-choice

(2AFC)

method,

(2)

The Japanese Psychonomic Society

NII-Electronic Library Service

The JapanesePsychonomic Society

116

The

Japanese

Journal

of

Psychonomic

Science

VoL27,

No,

1 Results

and

Discussion

Figure lshows

the

_performance as a function of

time,

Time

course data were

fitted

with

speed-accuracy trade-off

(SAT)

functions

(Eq.

1:

Carrasco

&

McElree, 20el), Performance increased monotoni-cally as duration increased

for

both

sentence

condi-tions.

Performance

for

the non-gaTden-path

sen-tences was

higher

than that

for

the _garden-path

sen-tences with

durations

longer

than

50

ms/phrase.

The

overall two-way within-subjects analysis of variance

{ANOVA)

revealed a significant interaction

between

sentence type and

duration

(F(5,

20)=4.42,

P<.O1),

with thesimple main effect of sentence type

significant forthe 16.7rnsfphrase

(F(1,

4)=8.0e,

P<

,05)

and 150 ms/phrase conditions

(F(1,

4)=9.41,

_P<

.05).

The simplc main effect of duration was

sig-nificant

for

both the non-garden-path and

garden-path conditions

(Fs{5,

20)>1O,P<.O1).

For

the

non-garden-path condition, multiple comparisons showed

thataccuracy forthe 16.7and 33.3 ms/phrase

condi-tions was lower than that

for

the other rates, and accuracy

for

the66.7rns/phrase condition was lower than thatforthe 150 and 250 ms!phrase conditions.

For the_garden-path condition, itwas also found that

accuracy

fer

the

16.7-66.7

rns/phrase conditions was

lower

than that forthe 100-250msfphrase

condi-tions;moreover, accuracy

for

the

100

and

150msl

phrase conditions was lower than that forthe 250

ms/phrase condition.

In

Eq.

1,

theparameterB isan

index

of thespeed of

processing. The estimated

fi

was 12.80 forthe

non-garden-path condition and

8.68 for

the garden-path

condition. This implies that the

_processing

of

gar-den-path

sentences was slower than that of

nen-garden-path sentences. The estimated _processing

times at the

85%

accuracy threshold were

104

ms

for

the non-garden-path condition and

174m$

tor

the

1.0 O.9 O.8 O.7 O.6 O.5o 100 200 Duration(ms!phrase)

Figure 1. Mean Accuracy

Error

bars

show

ISE.

300

F(t)=A<1-e-S"'a')

for

t>6else

_O

_(Eq.

₁₎

garden-path condition, There was a

difference

of

about

70

ms

between

the two conditions.

These resu}ts suggest that the processing time for

even simple _garden-path sentences

is

longer

than

for

non-garden-path sentences.

These

results indicateda

higher _processing cost for garden-path sentences in

Japanese.

The temporal cost was estimated

to

be

about

50-100

ms.

On

the other

hand,

_therewas no

di

fference

in

temporal cost

between

the two sentence

conditions at 250 msfphrase, probably

due

to a

ceil-ing effect. These results suggest _that _the

garden-path phenomenon

in

Japanese

did

not arise in

previ-eus studies because the _processing difference

be-tween garden-path and non-garden-path sentences only occurs with _phrases of relatively short duration.

References

Frazier,

L. &

Fodor,

J.

L.

(1978).

The sausage

chine] A new two-stage _parsing model. Cognition,6, 291-325.

Carrasco,

M,,

&

McElree,

B. (2001).

Convert

attention

accelerates the rate of visual

information

ing,

PiM4S, 98,5363-5367.