1. Dependency parsing (係り受け解析)
Naoaki Okazaki
okazaki at ecei.tohoku.ac.jp
#nlptohoku
Information Communication Theory (情報伝達学)

2. Information Communication Theory (情報伝達学)
Acknowledgements
Portions of this material are from:
Nivre and Kübler (2006). Dependency Parsing. Tutorial at Coling-ACL 2006.
長尾真編 (1996). 自然言語処理. 岩波講座ソフトウェア科学(15).
Information Communication Theory (情報伝達学)

3. Information Communication Theory (情報伝達学)
Dependency (依存, 係り受け)
Analyze syntactic structures of sentences using relationships between two words
sbj(loves, John): “John” is a nominal subject of the verb “loves”
obj(loves, Mary): “Mary” is a direct object of the verb “Mary”
Fit to languages with scrambling (語順の入れ替え)
root
sbj
obj
labeled or
unlabeled
(ROOT) John loves Mary .
type
head
dependent
governor
modifier
Information Communication Theory (情報伝達学)

4. Non-projective dependency tree
With crossing edges
John went a restaurant yesterday that served Sushi .
Information Communication Theory (情報伝達学)

5. Japanese dependency grammar (黒橋・長尾, 1996)
Dependency analysis has been explored extensively in Japanese, which allows scrambling (語順の入れ替え) and case ellipsis (格要素の省略)
Reversed direction of dependencies (traditionally)
Mostly right direction only
Exception:
Most Japanese written sentences are non-projective
私は　行きたくない　明日は
これが　僕は　正しいと　思う
Information Communication Theory (情報伝達学)

6. Arc-standard transitions
Three actions
Shift
Pop a token from the queue
Push the token into the stack
Left-arc
Pop a token from the stack
Add an edge
From: the top token in the queue
To: the popped token
Right-arc
From: the popped token
To: the top token in the queue
Stack
Queue [
a ... p ] [
q r … z ] [
a ... p q ] [
r … z ] [
a ... o p ] [
r … z ] [
a ... o ] [
r … z ] p [
a ... o p ] [
r … z ] [
a ... o ] [
p … z ] r
Information Communication Theory (情報伝達学)

7. Information Communication Theory (情報伝達学)
0: Initial state
Stack
Queue [
ROOT ] [
Economic
news
had
little
effect on
financial
markets . ]
Information Communication Theory (情報伝達学)

8. Information Communication Theory (情報伝達学)
1: Shift
Stack
Queue [
ROOT
Economic ] [
news
had
little
effect on
financial
markets . ]
Information Communication Theory (情報伝達学)

9. Information Communication Theory (情報伝達学)
2: Left-arc (nmod)
Stack
Queue [
ROOT ] [
news
had
little
effect on
financial
markets . ]
nmod
Economic
Information Communication Theory (情報伝達学)

10. Information Communication Theory (情報伝達学)
3: Shift
Stack
Queue [
ROOT
news ] [
had
little
effect on
financial
markets . ]
nmod
Economic
Information Communication Theory (情報伝達学)

11. Information Communication Theory (情報伝達学)
4: Left-arc (sbj)
Stack
Queue [
ROOT ] [
had
little
effect on
financial
markets . ]
sbj
news
nmod
Economic
Information Communication Theory (情報伝達学)

12. Information Communication Theory (情報伝達学)
5: Shift
Stack
Queue [
ROOT
had ] [
little
effect on
financial
markets . ]
sbj
news
nmod
Economic
Information Communication Theory (情報伝達学)

13. Information Communication Theory (情報伝達学)
6: Shift
Stack
Queue [
ROOT
had
little ] [
effect on
financial
markets . ]
sbj
news
nmod
Economic
Information Communication Theory (情報伝達学)

14. Information Communication Theory (情報伝達学)
7: Left-arc (nmod)
Stack
Queue [
ROOT
had ] [
effect on
financial
markets . ]
sbj
nmod
news
little
nmod
Economic
Information Communication Theory (情報伝達学)

15. Information Communication Theory (情報伝達学)
8: Shift
Stack
Queue [
ROOT
had
effect ] [ on
financial
markets . ]
sbj
nmod
news
little
nmod
Economic
Information Communication Theory (情報伝達学)

16. Information Communication Theory (情報伝達学)
9: Shift
Stack
Queue [
ROOT
had
effect on ] [
financial
markets . ]
sbj
nmod
news
little
nmod
Economic
Information Communication Theory (情報伝達学)

17. Information Communication Theory (情報伝達学)
10: Shift
Stack
Queue [
ROOT
had
effect on
financial ] [
markets . ]
sbj
nmod
news
little
nmod
Economic
Information Communication Theory (情報伝達学)

18. Information Communication Theory (情報伝達学)
10: Left-arc (nmod)
Stack
Queue [
ROOT
had
effect on ] [
markets . ]
sbj
nmod
nmod
news
little
financial
nmod
Economic
Information Communication Theory (情報伝達学)

19. Information Communication Theory (情報伝達学)
11: Right-arc (pmod)
Stack
Queue [
ROOT
had
effect ] [ on . ]
sbj
nmod
pmod
news
little
markets
nmod
nmod
Economic
financial
Information Communication Theory (情報伝達学)

20. Information Communication Theory (情報伝達学)
12: Right-arc (nmod)
Stack
Queue [
ROOT
had ] [
effect . ]
sbj
nmod
nmod
news
little on
nmod
pmod
Economic
markets
nmod
financial
Information Communication Theory (情報伝達学)

21. Information Communication Theory (情報伝達学)
13: Right-arc (obj)
Stack
Queue [
ROOT ] [
had . ]
sbj
obj
news
effect
nmod
nmod
nmod
Economic
little on
pmod
markets
nmod
financial
Information Communication Theory (情報伝達学)

22. Information Communication Theory (情報伝達学)
14: Shift
Stack
Queue [
ROOT
had ] [ . ]
sbj
obj
news
effect
nmod
nmod
nmod
Economic
little on
pmod
markets
nmod
financial
Information Communication Theory (情報伝達学)

23. Information Communication Theory (情報伝達学)
15: Right-arc (p)
Stack
Queue [
ROOT ] [
had ] p
sbj
obj
news .
effect
nmod
nmod
nmod
Economic
little on
pmod
markets
nmod
financial
Information Communication Theory (情報伝達学)

24. Information Communication Theory (情報伝達学)
16: Right-arc (root)
Stack
Queue [ ] [
ROOT ]
root
had p
sbj
obj
news .
effect
nmod
nmod
nmod
Economic
little on
pmod
markets
nmod
financial
Information Communication Theory (情報伝達学)

25. Information Communication Theory (情報伝達学)
17: Shift
Stack
Queue [
ROOT ] [ ]
root
had p
sbj
obj
news .
effect
nmod
nmod
nmod
Economic
little on
pmod
markets
nmod
financial
Information Communication Theory (情報伝達学)

26. Shift-reduce conflicts
What if we chose “Right-arc (obj)” at step 8?
We cannot attach the prepositional phrase on financial markets to effect (NP attachment) but only to had (VP attachment)!
Stack
Queue [
ROOT ] [
had on
financial
markets . ]
sbj
obj
news
effect
nmod
nmod
Economic
little
Information Communication Theory (情報伝達学)

27. Information Communication Theory (情報伝達学)
Summary
Two grammar formalisms and parsing algorithms
Constituency and dependency
The process of obtaining a tree from a sentence
(P)CKY and shift-reduce algorithm
Difficulty in parsing
Attachment and coordination
Time complexity and space requirement
Information Communication Theory (情報伝達学)