X hits on this document

PDF document

The split-fovea model of visual word - page 36 / 78

326 views

0 shares

0 downloads

0 comments

36 / 78

Chapter 2. The Simulations

31

’doze’ (which also appears in the training corpus) and hence will have a completely different meaning than the original input word ’dose’. Thus the network does not make a difference between the words ’dose’ and ’doze’.

coup

0.110

dose

0.109

dose

0.109

khan

0.107

Word Frequency Fixation Position Computed Output

Correct Output

all

kUUk

kUU

1,3,4

d@Uz

d@Us

2

dEUz

d@Us

1

JVrn

kVrn

Table 2.5: Errors with frequency 0.1 for the control network.

Word

Frequency

Fixation Position

Computed Output

Correct Output

coup

0.110

all

kUUJ

kUU

dose

0.109

1,2,3

d@Uz

d@Us

heir

0.110

all

I@r

E@r

hind

0.113

4

h&Vnd

h&Ind

hymn

0.106

1

gIm

hIm

khan

0.107

1

gVrn

kVrn

nude

0.104

4

njUUz

njUUd

shed

0.223

1

SVd

SEd

sigh

0.161

1,4

T&I

s&I

Table 2.6: Errors with frequency 0.1 for the fixation network. Note that here the

frequencies are the log frequencies as calculated from the CELEX database. Hence

there is a slight difference between the frequency shown and the actual frequency at

the different fixation positions.

Document info
Document views326
Page views340
Page last viewedSun Jan 22 15:25:00 UTC 2017
Pages78
Paragraphs3087
Words17414

Comments