Morphological disambiguation
by using machine learning
methods
Josef Bušta
12. 11. 2012
12. 11. 2012 Morphological disambiguation by using machine learning methods 2
Input examples (1)
Baseline: 93,5197
Total: 4 511 196
k3c4: 4 218 857
k7c7: 292 339
 Czes (465 102 710)
 Context: word1 word2 word3 class word4 word5 word6;
class 2 {k3c4, k7c7}
12. 11. 2012 Morphological disambiguation by using machine learning methods 3
Input examples (2)
Significant values ​​
of attribute (word4):
k1c7, k2c7,k3c7
Significant values ​​
of attribute (word5):
k1c7
12. 11. 2012 Morphological disambiguation by using machine learning methods 4
Input examples (3)
X: word5, Y: word4
word4 2 {k4c7, k4, k1c7, k3c7, k2c7} ! k7c7
12. 11. 2012 Morphological disambiguation by using machine learning methods 5
Results (1)
word1, word2, word3, word4, word5, word6, class
k1c1, k7c2, k1c2, k2c7, k1c7, kI, k7c7
k1c1, kI, k5, k1c2, k2c2, k1c2, k3c4
ZeroR J48 Id3 RF PART JRIp NB SMO IB3 BFTree
4500 93.5 98.2 96.95 98.2 98.3 96.9
1000 93.8 98.1 98.2
100 93.8 97.5 95.1 97.6 97.7 96.8
10 93.8 97 93.6 96.7 96.7 96.4 96.8 97.8 96.4 97.2
12. 11. 2012 Morphological disambiguation by using machine learning methods 6
Results (2)
* pos4, case4, pos5, case5, pos6, case6, class
k2, c7, k1, c7, kI, ?, k7c7
k1, c2, k2, c2, k1, c2, k3c4
J48 NB
4500* 97 96.2
4500** 97 96.1 ** word4, word5, word6, class
k2c7, k1c7, kI, k7c7
k1c2, k2c2, k1c2, k3c4
12. 11. 2012 Morphological disambiguation by using machine learning methods 7
Further work
 Inductive Logic Programming
 Data stream mining
 Testing different parameter settings for the algorithms