Improve Chinese Semantic Dependency Parsing via Syntactic Dependency
Parsing
Meishan Zhang
∗
, Wanxiang Che
∗
, Yanqiu Shao
†
and Ting Liu
∗
∗
Research Center for Social Computing and Information Retrieval
Harbin Institute of Technology, Harbin, China
Email: {mszhang,car,tliu}@ir.hit.edu.cn
†
Artificial Intelligence Institute
Beijing City University , Beijing, China
Email: yqshao@bcu.edu.cn
Abstract—We address the problem of Chinese semantic
dependency parsing. Dependency parsing is traditionally
oriented to syntax analysis, which we denote by syntactic
dependency parsing to distinguish it from semantic depen-
dency parsing. In this paper, firstly we compare Chinese
semantic dependency parsing and syntactic dependency
parsing systematically, showing that syntactic dependency
parsing can potentially improve the performance of seman-
tic dependency parsing. Thus then we suggest an approach
based on quasi-synchronous grammar to incorporate the
auto-parsed syntactic dependency tree into semantic de-
pendency parsing. We conduct experiments on the Chinese
semantic dependency parsing corpus of SemEval-2012.
Finally we achieve 65.25% LAS on test corpus, gaining
increases of 2.45% compared to the top result of 62.80%
in SemEval-2012.
Keywords-Semantic Dependency Parsing, Syntactic De-
pendency Parsing, Quasi-synchronous Grammar.
I. INTRODUCTION
Chinese semantic dependency parsing has been pro-
posed to facilitate semantic analysis for a long time
[1]. It is supposed that semantic dependency parsing
can be useful for many applications such as question
answering, textual entailment, machine translation, in-
formation extraction, etc. Relatively less work has been
done on it. This could be due to two facts. First, there
isn’t a unified standard since linguistic phenomenons
are too difficult to catch comprehensively. Second, there
isn’t a large scale annotated corpus which is available
for public . Fortunately, SemEval-2012 has organized a
task for Chinese semantic dependency parsing [ 2]. Nine
systems have been submitted. The top result achieves
62.80%
1
in labeled attachment score (LAS) and 80.45%
in unlabeled attachment score (UAS).
In this paper, we suggest to enhance Chinese semantic
dependency parsing by syntactic dependency parsing.
Seldom works have considered the syntax constraints for
Chinese semantic dependency parsing. We expect better
performance would be gained if syntactic dependency
parsing is done preliminarily. Firstly, we analyze the
similarities and differences of the two different depen-
dency parsing problems. We find syntactic dependency
1
Some teams have renewed their results after acquiring the test
corpus.
parsing can potentially help semantic dependency pars-
ing. Thus secondly we attempt to make use of syntactic
dependency parsing for semantic dependency parsing.
We adopt the approach mentioned in [3], which gen-
erates rich quasi-synchronous grammar (QG) features
generated by transformation patterns (TP). Experiments
are conducted on corpus of [2]. Our baseline system
employs Mate [4], achieving a LAS of 64.65% which is
1.85% higher than the top result in SemEval-2012. Then
we use the corpus of CONLL 2009 [5] to train a syn-
tactic dependency parsing model. Guided by results of
syntactic dependency parsing, our final system achieves
a LAS of 65.25%.
The remainder of the paper is organized as follows.
Section 2 gives a comparison between Chinese semantic
dependency parsing and syntactic dependency parsing.
Section 3 describes our system for Chinese semantic
dependency parsing. Section 4 reports the experimental
results and error analysis. Section 5 describes the related
works. Finally, in Section 6 we conclude this paper.
II. C
HINESE SEMANTIC DEPENDENCY PARSING:A
C
OMPARISON WITH SYNTACTIC DEPENDENCY PARSING
In this section, we analyze the similarities and dif-
ferences of Chinese semantic dependency parsing and
syntactic dependency parsing by comparing the corpus
of [2] and CONLL 2009. Figure 1 shows an example
of the two dependency parsing problems. The red part
displays the dependency tree of semantic dependency
parsing, and the blue part displays the tree of syntactic
dependency parsing.
㒣ᐌ
Always
㹿
bei
В՟
illustration
ⱘ
de
ᰃ
is
㑶♃
Hongdeng Ye
ⱘ
de
ϰ㘮
Dong-ju
⬉Ϯ
electric company
frequency
prep-depend aux-depend
d-genetive
d-member
isa
ROOT
relevant
aux-depend
AUX
ADV
COMP SBJ
COMP
NMOD
NMOD
PRDROOT
Figure 1. A comparison example between Chinese semantic depen-
dency parsing and syntactic dependency parsing.
2012 International Conference on Asian Language Processing
978-0-7695-4886-9/12 $26.00 © 2012 IEEE
DOI 10.1109/IALP.2012.42
53