X hits on this document

PDF document

The split-fovea model of visual word - page 41 / 78

242 views

0 shares

0 downloads

0 comments

41 / 78

Fixation Position

Control Network Fixation Network

1

29

24

2

23

27

3

24

27

4

30

25

5

33

33

Table 2.9: Errors by fixation position. The last column, labelled

difference in errors between the control and the fixation network

errors of control).

Fixation Position

Control Network Fixation Network

1

0.177

0.199

2

0.166

0.212

3

0.177

0.202

4

0.179

0.187

5

0.170

0.190

Table 2.10: Mean Sum Sq

uared Error for the whole training set by fixation position.

Again the last column gives

the difference in mse between the two networks.

Chapter 2. The Simulations

Change

+0.022 +0.046 +0.025 +0.008 +0.020

36

Change

  • -

    4

+4 +3

  • -

    5

+0

’change’ gives the

(errors of fixation -

beginning of English words are less predictable and thus harder to learn than the more predictable ending of words. Thus the higher error rate for the onset versus the coda.

2.2.5

Errors by Fixation Position

An interesting statistic for comparing the two different networks to be able to tell if they actually behave differently is to look at how the networks behave at the different fixation positions. This has been tested in two different ways: table 2.9 shows the number of errors the networks made at the different fixation positions while table 2.10

Document info
Document views242
Page views251
Page last viewedSun Dec 11 04:59:27 UTC 2016
Pages78
Paragraphs3087
Words17414

Comments