Download ppt - Deniz Yuret

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Document related concepts
no text concepts found
Transcript
Dependency Parsing as a
Classification Problem
Deniz Yuret
Koç University
İstanbul
Linking Adjacent Words
kick the red ball
?
?
?
Linking Adjacent Words
kick the red ball
Linking Adjacent Words
kick the red ball
?
?
Linking Adjacent Words
kick the red ball
?
?
?
Linking Adjacent Words
kick the red ball
Linking Adjacent Words
kick the red ball
?
?
?
?
Linking Adjacent Words
kick the red ball
?
?
?
Linking Adjacent Words
kick the red ball
?
?
Linking Adjacent Words
kick the red ball
Percentage of adj words linked
Language
% adjacent
words linked
Language
% adjacent
words linked
Arabic
61.02
Japanese
54.81
Chinese
56.59
Portuguese 50.81
Czech
48.73
Slovene
45.62
Danish
55.93
Spanish
51.28
Dutch
55.54
Swedish
48.26
German
42.15
Turkish
62.60
Sample decision list for German
1. If
Then
2. If
Then
3. If
Then
4. If
Then
XL1:postag=APPR
NONE
X:postag=ART, Y:postag=NN
L:NK
X:postag=APPR
R:NK
TRUE
NONE
Sample decision list for German
1. If
Then
2. If
Then
3. If
Then
4. If
Then
XL1:postag=APPR
NONE
X:postag=ART, Y:postag=NN
L:NK
X:postag=APPR
R:NK
TRUE
NONE
APPR-ART-NN
Attribute Selection
Language
Attributes
Language
Attributes
Arabic
ALL
Japanese
postag,suffix2
Chinese
postag,cpostag
Portuguese postag,lemma
Czech
postag,lemma
Slovene
ALL
Danish
postag,form
Spanish
postag,lemma
Dutch
postag,feats
Swedish
postag,form
German
postag,form
Turkish
ALL
Accuracy on adj word links
Language
% adjacent
words linked
% adjacent
link accuracy
Language
% adjacent
words linked
% adjacent
link accuracy
Arabic
61.02
76.87
Japanese
54.81
95.56
Chinese
56.59
84.51
Portuguese 50.81
90.18
Czech
48.73
79.25
Slovene
45.62
85.19
Danish
55.93
86.96
Spanish
51.28
89.01
Dutch
55.54
85.36
Swedish
48.26
83.20
German
42.15
87.97
Turkish
62.60
85.27
Contributions
• Learning adjacent word dependencies
using decision lists and GPA.
• Accuracy on adjacent word dependencies
between 85%-95% for most languages.
• Greedy bottom-up parsing model did not
work very well.
Related documents