Proceedings of the Human Language Technology Conference of the North American Chapter of the ACL, pages 431–438,
New York, June 2006. c©2006 Association for Computational Linguistics
Semantic Role Labeling of Nominalized Predicates in Chinese
Nianwen Xue
Center for Research in Spoken Language
University of Colorado
Boulder, CO, 80309
Nianwen.Xue@colorado.edu
Abstract
Recent work on semantic role labeling
(SRL) has focused almost exclusively on
the analysis of the predicate-argument
structure of verbs, largely due to the lack
of human-annotated resources for other
types of predicates that can serve as train-
ing and test data for the semantic role
labeling systems. However, it is well-
known that verbs are not the only type
of predicates that can take arguments.
Most notably, nouns that are nominalized
forms of verbs and relational nouns gen-
erally are also considered to have their
own predicate-argument structure. In this
paper we report results of SRL experi-
ments on nominalized predicates in Chi-
nese, using a newly completed corpus,
the Chinese Nombank. We also dis-
cuss the impact of using publicly avail-
able manually annotated verb data to im-
prove the SRL accuracy of nouns, exploit-
ing a widely-held assumption that verbs
and their nominalizations share the same
predicate-argument structure. Finally, we
discuss the results of applying reranking
techniques to improve SRL accuracy for
nominalized predicates, which showed in-
significant improvement.
1 Introduction
Detecting and classifying the arguments of predi-
cates has been an active area of research in recent
years, driven by the availability of large-scale se-
mantically annotated corpora such as the FrameNet
(Baker et al., 1998) and the Propbank (Palmer et
al., 2005). It is generally formulated as a seman-
tic role labeling (SRL) task, where each argument
of the predicate is assigned a label that represents
the semantic role it plays with regard to its pred-
icate (Gildea and Jurafsky, 2002; Hacioglu et al.,
2003; Pradhan et al., 2004b; Xue and Palmer, 2004;
Toutanova et al., 2005; Koomen et al., 2005). It
has been the shared task for the CoNLL competition
for two consecutive years (Carreras and M`arquez,
2004b; Carreras and M`arquez, 2005). This line of
research has also expanded from English to other
languages (Sun and Jurafsky, 2004; Xue and Palmer,
2005). So far, however, most of the research efforts
have focused on analyzing the predicate-argument
structure of verbs, largely due to absence of an-
notated data for other predicate types. In this pa-
per, we report SRL experiments performed on nom-
inalized predicates in Chinese, taking advantage of
a newly completed corpus, the Chinese Nombank
(Xue, 2006), which we describe in greater detail in
Section 2. The rest of the paper is organized as fol-
lows. Section 3 describes the architecture of our sys-
tem as well as the features we used in our experi-
ments. In Section 4 we describe the experimental
setups and report our experimental results. We first
present experiments that use hand-crafted parses as
input, providing a measurement of how well the
Nombank annotation can be bootstrapped from the
syntactic structure in the treebank. We then describe
a more realistic experimental setup in which an au-
tomatic parser is first used to parse unsegmented raw
431
text and its output is then fed into our SRL system.
We also discuss whether verb data can be used to im-
prove the SRL accuracy of nominalized predicates.
Finally we describe a preliminary experiment that
uses reranking techniques to improve the SRL ac-
curacy on hand-crafted parses. Section 5 attempts to
put our results in perspective in the context of related
work. Section 6 concludes our paper.
2 The Chinese Nombank
The Chinese Nombank extends the general anno-
tation framework of the English Proposition Bank
(Palmer et al., 2005) and the English Nombank
(Meyers et al., 2004) to the annotation of nomi-
nalized predicates in Chinese. Like the English
Nombank project, the Chinese Nombank adds a
layer of semantic annotation to the Chinese Tree-
Bank (CTB), a syntactically annotated corpus of 500
thousand words. The Chinese Nombank annotates
two types of elements that are associated with the
nominalized predicate: argument-like elements that
are expected of this predicate, and adjunct-like el-
ements that modify this predicate. Arguments are
assigned numbered labels (prefixed by ARG, e.g.,
ARG0...ARGn) while adjuncts receive a functional
tag (e.g., TMP for temporal, LOC for locative, MNR
for manner) prefixed by ARGM. A predicate gen-
erally has no more than six numbered arguments
and the complete list of functional tags for adjuncts
and their descriptions can be found in the annotation
guidelines of this project.
The Chinese Nombank also adds a coarse-grained
sense tag to the predicate. The senses of a predicate,
formally called framesets, are motivated by the ar-
gument structure of this predicate and are thus an
integral part of the predicate-argument structure an-
notation. Sense disambiguation is performed only
when different senses of a predicate require different
sets of arguments. These senses are the same senses
defined for the corresponding verbs in the Chinese
Proposition Bank, but typically only a subset of the
verb senses are realized in their nominalized forms.
The example in 1 illustrates the Chinese Nombank
annotations, which are the labels in bold in the parse
tree. Take cjkB7A2cjkD5B9(”development”) as an example, f1
is the frameset identifier. Of the four expected argu-
ments for this frameset, ARG0 the cause or agent,
ARG1 the theme, ARG2 the initial state and ARG3
the end state or goal, only ARG1 is realized and it
is cjkC1BDcjkB0B6cjkB9D8cjkCFB5(”cross-Strait relations”). The predi-
cate also has a modifier labeled ARGM-TMP, cjkBDF1
cjkBAF3(”hereafter”).
Typically the arguments and adjuncts of a nomi-
nalized predicate are realized inside the noun phrase
headed by the nominalized predicate, as is the case
for cjkB7A2cjkD5B9(“development”) in Example 1. A main
exception is when the noun phrase headed by the
nominalized predicate is an object of a support verb,
in which case the arguments of this predicate can
occur outside the noun phrase. This is illustrated
by cjkB9E6cjkBBAE(“planning”) in Example 1, where the noun
phrase of which it is the head is the object of a sup-
port verb cjkBDF8cjkD0D0(“conduct”), which has little mean-
ing of its own. Both arguments of this predicate,
cjkBAA3cjkCFBFcjkC1BDcjkB0B6(“the two sides of the Taiwan Strait”)
and cjkBDF1cjkBAF3cjkC1BDcjkB0B6cjkB9D8cjkCFB5cjkB5C4cjkB7A2cjkD5B9(“the development of
the cross-Strait relations”), are realized outside the
noun phrase. There are also a few other general ten-
dencies about the arguments of nominalized predi-
cates that are worth pointing out. The distribution of
their arguments is much less predictable than verbs
whose arguments typically occupy prominent syn-
tactic positions like the subject and object. There
also tend to be fewer arguments that are actually
realized for nominalized predicates. Nominalized
predicates also tend to take fewer types of adjuncts
(ARGMs) than their verbal counterpart and they
also tend to be less polysemous, having only a subset
of the senses of their verb counterpart.
The goal of the semantic role labeling task de-
scribed in this paper is to identify the arguments
and adjuncts of nominalized predicates and assign
appropriate semantic role labels to them. For the
purposes of our experiments, the sense information
of the predicates are ignored and left for future re-
search.
3 System description
The predominant approach to the semantic role la-
beling task is to formulate it as a classification prob-
lem that can be solved with machine-learning tech-
niques. Argument detection is generally formulated
as a binary classification task that separates con-
stituents that are arguments or adjuncts to a pred-
432
IP
NP-SBJ VP
ARG0/REL2 VV VP
cjkBAA3cjkCFBFcjkC1BDcjkB0B6
the two sides
of the Straits
cjkBFC9
can
PP-DIR VP
ARG1/REL2 VV NN
P NP SUP/REL2 REL2
cjkB6D4
regarding
NP DNP NP cjkBDF8cjkD0D0
conduct
f1
ARGM-TMP/REL1 NP DEG NN cjkB9E6cjkBBAE
plan
cjkBDF1cjkBAF3
hereafter
NP cjkB5C4 REL1
ARG1/REL1 f1
cjkC1BDcjkB0B6cjkB9D8cjkCFB5
Cross-Strait
relations
cjkB7A2cjkD5B9
development
The two sides of the Taiwan Straits can plan the development of the cross-Strait relations hereafter.
Table 1: A nominalized predicate annotated with semantic roles
icate from those that are not related to the pred-
icate in question. Argument classification, which
classifies the constituents into a category that cor-
responds to one of the argument or adjunct la-
bels is a natural multi-category classification prob-
lem. Many classification techniques, SVM (Pradhan
et al., 2004b), perceptrons (Carreras and M`arquez,
2004a), Maximum Entropy (Xue and Palmer, 2004),
etc. have been successfully used to solve SRL prob-
lems. For our purposes here, we use a Maximum En-
tropy classifier with a tunable Gaussian prior in the
Mallet Toolkit1. The Maximum Entropy classifier
does multi-category classification and thus can be
1http://mallet.cs.umass.edu
straightforwardly applied to the problem here. The
classifier can be tuned to minimize overfitting by ad-
justing the Gaussian prior.
3.1 A three-stage architecture
Like verbal predicates, the arguments and adjuncts
of a nominalized predicate are related to the pred-
icate itself in linguistically well-understood struc-
tural configurations. As we pointed out in Section
2, most of the arguments for nominalized predicates
are inside the NP headed by the predicate unless the
NP is the object of a support verb, in which case its
arguments can occur outside the NP. Typically the
subject of the support verb is also an argument of the
nominalized predicate, as illustrated in Example 1.
433
The majority of the constituents are not related to the
predicate in question, especially since the sentences
in the treebank tend to be very long. This is clearly
a lingustic observation that can be exploited for the
purpose of argument detection. There are two com-
mon approaches to argument detection in the SRL
literature. One is to apply a binary classifier directly
to all the constituents in the parse tree to separate
the arguments from non-arguments, and let the ma-
chine learning algorithm do the work. This can be
done with high accuracy when the machine-learning
algorithm is powerful and is provided with appro-
priate features (Hacioglu et al., 2003; Pradhan et
al., 2004b). The alternative approach is to combine
heuristic and machine-learning approaches (Xue and
Palmer, 2004). Some negative samples are first fil-
tered out with heuristics that exploit the syntactic
structures represented in a parse tree before a binary
classifier is applied to further separate the positive
samples from the negative samples. It turns out the
heuristics that are first proposed in Xue and Palmer
(2004) to prune out non-arguments for verbal pred-
icates can be easily adapted to detect arguments for
the nominalized predicates as well, so in our exper-
iments we adopt the latter approach. The algorithm
starts from the predicate that anchors the annotation,
and first collects all the sisters of this predicate. It
then iteratively moves one level up to the parent of
the current node to collect its sisters till it reaches the
appropriate top-level node. At each level, the sys-
tem has a procedure to determine whether that level
is a coordination structure or a modification struc-
ture. The system only considers a constituent to be
a potential candidate if it is an adjunct to the current
node. Punctuation marks at all levels are skipped.
After this initial procedure, a binary classifier is ap-
plied to distinguish the positive samples from the
negative samples. A lower threshold is used for pos-
itive samples than negative samples to maximize the
recall so that we can pass along as many positive
samples as possible to the next stage, which is the
multi-category classification.
3.2 Features
SRL differs from low-level NLP tasks such as POS
tagging in that it has a fairly large feature space and
as a result linguistic knowledge is crucial in design-
ing effective features for this task. A wide range of
features have been shown to be useful in previous
work on semantic role labeling for verbal predicates
(Gildea and Jurafsky, 2002; Pradhan et al., 2004b;
Xue and Palmer, 2004) and our experiments show
most of them are also effective for SRL of nominal-
ized predicates. The features for our multicategory
classifier are listed below:
• Predicate: The nominalized predicate itself.
• Position: The position is defined in relation to
the predicate and the values are before and af-
ter. Since most of the arguments for nominal-
ized predicates in Chinese are before the predi-
cates, this feature is not as effective as when it
is used for verbal predicates.
• path: The path between the constituent being
classified and the predicate.
• path+dominating verb. The path feature com-
bined with the dominating verb. This feature is
only invoked when there is an intervening dom-
inating verb between the constituent being clas-
sified and the predicate. It is used to capture
the observation that only a closed set of verbs
can be support verbs for nominalized predicates
and they are good indicators of whether or not
the constituent is an argument of this predicate
and the semantic role of the argument.
• Head word and its part of speech: The head
word and its part-of-speech have proved to be
a good indicator of the semantic role label of
a constituent for verbal predicates in previous
work. It proves to be a good feature for nominal
predicates as well.
• Phrase type: The syntactic category of the con-
stituent being classified.
• First and last word of the constituent being
classified
• sisterhood with predicate: A binary feature that
indicates whether the constituent being classi-
fied is a sister to the nominalized predicate.
• Combination features: predicate-head word
combination, predicate-phrase type combina-
tion.
434
• class features. Features that replace the pred-
icate with its class. The class features are in-
duced from frame files through a procedure first
introduced in (Xue and Palmer, 2005).
Not all the features used for multicategory clas-
sification are equally effective for binary classifica-
tion, which only determines whether or not a con-
stituent is an argument or adjunct to the nominal-
ized predicate. Therefore, the features for the binary
classifier are a subset of the features used for multi-
category classification. These are path, path plus
dominating verb, head word and its part-of-speech
and sisterhood.
4 Experiments
4.1 Data
Our system is trained and tested on a pre-release
version of the Chinese Nombank. This version of
the Chinese Nombank consists of standoff annota-
tion on the first 760 articles (chtb_001.fid to
chtb_931.fid) of the Penn Chinese Treebank2.
This chunk of data has 250K words and 10,364 sen-
tences. It has 1,227 nominalized predicate types and
10,497 nominalized predicate instances. In com-
parison, there are 4,854 verb predicate types and
37,183 verb predicate instances in the same chunk
of data. By instance, the size of the Nombank is be-
tween a quarter and one third of the Chinese Propo-
sition Bank. Following the convention of the se-
mantic role labeling experiments in previous work,
we divide the training and test data by the num-
ber of articles, not by the predicate instances. This
pretty much guarantees that there will be unseen
predicates in the test data. For all our experiments,
688 files are used as training data and the other
72 files (chtb_001.fidto chtb_040.fidand
chtb_900.fidto chtb_931.fid) are held out
as test data. The test data is selected from the
double-annotated files in the Chinese Treebank and
the complete list of double-annotated files can be
found in the documentation for the Chinese Tree-
bank 5.1. Our parser is trained and tested with the
same data partition as our semantic role labeling sys-
tem.
2The most current version (CTB5.1) of the Penn Chinese
Treebank has 507K words, 825K Chinese characters, 18,716
sentences and 890 articles.
4.2 Semantic role tagging with hand-crafted
parses
In this section we present experimental results us-
ing Gold Standard parses in the Chinese Treebank
as input. To be used in real-world natural language
applications, a semantic role tagger has to use au-
tomatically produced constituent boundaries either
from a parser or by some other means, but experi-
ments with Gold Standard input will help us evaluate
how much of a challenge it is to map a syntactic rep-
resentation to a semantic representation, which may
very well vary from language to language. There
are two experimental setups. In the first experiment,
we assume that the constituents that are arguments
or adjuncts are known. We only need to assign the
correct argument or adjunct labels. In the second
experiment, we assume that all the constituents in a
parse tree are possible arguments. The system first
filters out consituents that are highly unlikely to be
an argument for the predicate, using the heuristics
described in Section 3. A binary classifier is then
applied to the remaining constituents to do further
separation. Finally the multicategory classifier is
applied to the candidates that the binary classifier
passes along. The results of these two experiments
are presented in Table 2.
experiments all corep (%) r(%) f(%) f(%)
constituents known n/a n/a 86.6 86.9
constituents unknown 69.7 73.7 71.6 72.0
Table 2: Results for hand-crafted parses
Compared with the 93.9% reported by Xue and
Palmer (2005) for verbal predicates on the same
data, the 86.9% the system achieved when the con-
situents are given is considerably lower, suggest-
ing that SRL for nominalized predicates is a much
more challenging task. The difference between the
SRL accuracy for verbal and nominalized predicates
is even greater when the constituents are not given
and the system has to identify the arguments to be
classified. Xue and Palmer reported an f-score of
91.4% for verbal predicates under similar experi-
mental conditions, in contrast with the 71.6% our
system achieved for nominalized predicates. Care-
ful error analysis shows that one important cause for
435
this degradation in performance is the fact that there
is insufficient training data for the system to reliably
separate support verbs from other verbs and deter-
mine whether the constituents outside the NP headed
by the nominalized predicate are related to the pred-
icate or not.
4.3 Using automatic parses
We also conducted an experiment that assumes a
more realistic scenario in which the input is raw un-
segmented text. We use a fully automatic parser
that integrates segmentation, POS tagging and pars-
ing. Our parser is similar to (Luo, 2003) and is
trained and tested on the same data partition as the
semantic role labeling system. Tested on the held-
out test data, the labeled precision and recall are
83.06% and 80.15% respectively for all sentences.
The results are comparable with those reported in
Luo (Luo, 2003), but they cannot be directly com-
pared with most of the results reported in the litera-
ture, where correct segmentation is assumed. In ad-
dition, in order to account for the differences in seg-
mentation, each character has to be treated as a leaf
of the parse tree. This is in contrast with word-based
parsers where words are terminals. Since semantic
role tagging is performed on the output of the parser,
only constituents in the parse tree are candidates. If
there is no constituent in the parse tree that shares
the same text span with an argument in the manual
annotation, the system cannot possibly get a correct
annotation. In other words, the best the system can
do is to correctly label all arguments that have a con-
stituent with the same text span in the parse tree.
all core
p (%) r(%) f(%) f(%)
49.7 53.1 51.3 48.3
Table 3: Results for automatic parses
The results show a similar performance degrada-
tion compared with the results reported for verbs on
the same data in previous work, which is not unex-
pected. Xue and Palmer (2005) reported an f-score
of 61.3% when a parser is used to preprocess the
data.
4.4 Using verb data to improve noun SRL
accuracy
Since verbs and their nominalized counterparts are
generally considered to share the same argument
structure and in fact the Chinese Nombank is an-
notated based on the same set of lexical guide-
lines (called frame files) as the Chinese PropBank,
it seems reasonable to expect that adding the verb
data to the training set will improve the SRL accu-
racy of the nominal predicates, especially when the
training set is relatively small. Given that verbs and
their nominalized counterpart share the same mor-
phological form in Chinese, adding the verb data to
the training set is particularly straightforward. In
our experiments, we extracted verb instances from
the CPB that have nominalized forms in the portion
of the Chinese Treebank on which our SRL exper-
iments are performed and added them to the train-
ing set. Our experiments show, however, that sim-
ply adding the verb data to the training set and in-
discriminately extracting the same features from the
verb and noun instances will hurt the overall perfor-
mance instead of improving it. This result is hardly
surprising upon closer examination: the values of
certain features are vastly different for verbal and
nominal predicates. Most notably, the path from the
predicate to the constituent being classified, an im-
portant feature for semantic role labeling systems,
differ greatly from nominal and verbal predicates.
When they are thrown in the same training data mix,
they effectively create noise and neutralize the dis-
criminative effect of this feature. Other features,
such as the head words and their POS tags, are the
same and adding these features does indeed improve
the SRL accuracy of nominal predicates, although
the improvement is not statistically significant.
4.5 Reranking
In a recent paper on the SRL on verbal predicates
for English, (Toutanova et al., 2005) pointed out that
one potential flaw in a SRL system where each ar-
gument is considered on its own is that it does not
take advantage of the fact that the arguments (not the
adjuncts) of a predicate are subject to the hard con-
straint that they do not have the same label3. They
3For certain symmetrical predicates, arguments can have the
same label, although these cases are rare.
436
show that by performing joint learning of all the ar-
guments in the same proposition (for the same predi-
cate), the SRL accuracy is improved. To test the effi-
cacy of joint-learning for nominalized predicates in
Chinese, we conducted a similar experiment, using
a perceptron reranker described in Shen and Joshi
(2004). Arguments and adjuncts of the same predi-
cate instance (proposition) are chained together with
their joint probability being the product of the indi-
vidual arguments and the top K propositions are se-
lected as the reranking candidates. When the argu-
ments are given and the input is hand-crafted gold-
standard parses in the treebank, selecting the top 10
propositions yields an oracle score of 97%. This ini-
tial promise does not pan out, however. Performing
reranking on the top 10 propositions did not lead
to significant improvement, using the five feature
classes described in (Haghighi et al., 2005). These
are features that are hard to implement for individual
arguments: core argument label sequence, flattened
core argument label sequence, core argument labels
and phrase type sequence, repeated core argument
labels with phrase types, repeated core argument la-
bels with phrase types and adjacency information.
We speculate that the lack of improvement is due
to the fact that the constraint that core (numbered)
arguments should not have the same semantic role
label for Chinese nominalized predicates is not as
rigid as it is for English verbs. However further error
analysis is needed to substantiate this speculation.
5 Related Work
Compared with large body of work on the SRL
of verbal predicates, there has been relatively lit-
tle work done in analyzing the predicate-argument
structure of nominalized predicates. There are even
less work done for the nominalized predicates for
Chinese. (Hull and Comez, 1996) implemented a
rule-based system for identifying the arguments for
nominal predicates and (Lapata, 2002) has a system
that interprets the relation between the head of noun
compound and its head, but no meaningful compar-
ison can be made between our work and theirs. Per-
haps the closest work to that of ours is that of (Prad-
han et al., 2004a), where they reported preliminary
work for analyzing the predicate-argument structure
of Chinese nominalizations, using a small data set of
630 proposition for 22 nominalizations taken from
the Chinese Treebank. Since different data sets are
used, the results cannot be meaningfully compared.
The results reported here for nominalized pred-
icates are consistent with what Xue and Palmer
(2005) reported for the SRL of Chinese verbs with
regard to the role of the parser in their semantic
role labeling system: there is a substantial perfor-
mance drop when the automatic parser is used. At
present, improvement in Chinese parsing is hindered
by insufficient training material. Although the Chi-
nese Treebank has a decent size of 500K words, it
is evenly divided into two portions of very differ-
ent sources, Xinhua newswire from mainland China
and Sinorama magazines from Taiwan. Due to their
very different styles, training on one portion of the
data does not help or even hurt the parsing accuracy
of the other portion. The lack of sufficient train-
ing material is compounded by inherent properties
of the Chinese language that makes Chinese pars-
ing particularly difficult. Chinese segmentation is
a much more difficult problem than tokenization of
English text and Chinese words do not have mor-
phological clues that can help parsing decisions. We
believe further improvement in SRL accuracy will
be to a large extent contingent on the parsing accu-
racy, which requires more training material.
6 Conclusion and future work
We reported first results on the semantic role label-
ing of nominalized predicates in Chinese, using a
sizable annotated corpus, the Chinese Nombank, as
training and test data. Compared with that of ver-
bal predicates, SRL of nominalized predicates gen-
erally presents a more challenging problem, for all
experimental conditions. While the smaller train-
ing set compared with that of verbal predicates may
provide partial explanation for the degradation in
performance, we believe another important reason
is that the arguments for nominalized predicates do
not occupy prominent syntactic positions such as the
subject and object, as arguments of verbal predicates
often do. As a result, the syntactic structure repre-
sented in the parse tree does not provide as much of a
clue for their detection and classification. However,
this makes SRL of nominalized predicates a more
pressing issue to solve, as they represent a substan-
437
tial proportion of the predicates in the corpus. Our
results also show that the k-best propositions pro-
duced by the local classifier have a very high ora-
cle score, which perhaps indicates a promising path
that deserves further exploration, based on careful
analysis of the errors. We intend to continue to ex-
periment with new features and parameters for the
reranking algorithm.
7 Acknowledgement
I would like to thank Martha Palmer for her unwa-
vering support for this line of research. This work
is funded by the NSF ITR via grant 130-1303-4-
541984-XXXX-2000-1070.
References
Collin F. Baker, Charles J. Fillmore, and John B. Lowe.
1998. The Berkeley FrameNet project. In Pro-
ceedings of COLING/ACL, pages 86–90, Montreal,
Canada.
Xavier Carreras and Llu´ıs M`arquez. 2004a. Hierarchi-
cal Recognition of Propositional Arguments with Per-
ceptrons. In Proceedings of the Eighth Conference on
Natural Language Learning, Boston, Massachusetts.
Xavier Carreras and Llu´ıs M`arquez. 2004b. Introduction
to the CoNLL-2004 Shared Task: Semantic Role La-
beling. In Proceedings of the Eighth Conference on
Natural Language Learning, Boston, Massachusetts.
Xavier Carreras and Llu´ıs M`arquez. 2005. Introduction
to the CoNLL-2005 Shared Task: Semantic Role La-
beling. In Proceedings of the Nineth Conference on
Natural Language Learning, Ann Arbor, Michigan.
D. Gildea and D. Jurafsky. 2002. Automatic label-
ing for semantic roles. Computational Linguistics,
28(3):245–288.
Kadri Hacioglu, Sameer Pradhan, Wayne Ward, James H.
Martin, and Daniel Jurafsky. 2003. Shallow Seman-
tic Parsing Using Support Vector Machines. Technical
Report CSLR-2003-1, Center for Spoken Language
Research at the University of Colorado.
Aria Haghighi, Kristina Toutanova, and Christopher
Manning. 2005. A Joint Model for Semantic Role
Labeling. In Proceedings of the Nineth Conference on
Natural Language Learning, Ann Arbor, Michigan.
Richard D. Hull and Fernando Comez. 1996. Semantic
interpretation of nominalizations. In The AAAI Con-
ference, pages 1062–1068, Oregon.
Peter Koomen, Vasin Punyakanok, Dan Roth, and Wen
tau Yih. 2005. Generalized Inference with Multiple
Semantic Role Labeling Systems. In Proceedings of
the Nineth Conference on Natural Language Learning,
Ann Arbor, Michigan.
Maria Lapata. 2002. The disambiguation of nominaliza-
tions. Computational Linguistics, 28(3):357–388.
Xiaoqiang Luo. 2003. A Maximum Entropy Chinese
Character-Based Parser. In Proceedings of the 2003
Conference on Empirical Methods in Natural Lan-
guage Processing (EMNLP 2003), Sapporo, Japan.
A. Meyers, R. Reeves, C. Macleod, R. Szekely, V. Zielin-
ska, B. Young, and R. Grishman. 2004. The Nom-
Bank Project: An Interim Report. In Proceedings of
the NAACL/HLT Workshop on Frontiers in Corpus An-
notation, Boston, Massachusetts.
Martha Palmer, Dan Gildea, and Paul Kingsbury. 2005.
The Proposition Bank: An Annotated Corpus of Se-
mantic Roles. Computational Linguistics, 31(1).
Sameer Pradhan, Honglin Sun, Wayne Ward, James H.
Martin, and Daniel Jurafsky. 2004a. Parsing Argu-
ments of Nominalizations in English and Chinese. In
Proceedings of NAACL-HLT 2004, Boston, Mass.
Sameer Pradhan, Wayne Ward, Kadri Hacioglu, James H.
Martin, and Daniel Jurafsky. 2004b. Shallow Seman-
tic Parsing Using Support Vector Machines. In Pro-
ceedings of NAACL-HLT 2004, Boston, Mass.
Libin Shen and Aravind K. Joshi. 2004. Flexible Margin
Selection for Reranking with Full Pairwise Samples.
In Proceedings of IJCNLP-2004, pages 446–455.
Honglin Sun and Daniel Jurafsky. 2004. Shallow Se-
mantic Parsing of Chinese. In Proceedings of NAACL
2004, Boston, USA.
Kristina Toutanova, Aria Haghighi, and Christopher
Manning. 2005. Joint Learning Improves Semantic
Role Labeling. In Proceedings of ACL-2005.
Nianwen Xue and Martha Palmer. 2004. Calibrating
features for Semantic Role Labeling. In Proceedings
of 2004 Conference on Empirical Methods in Natural
Language Processing, Barcelona, Spain.
Nianwen Xue and Martha Palmer. 2005. Automatic Se-
mantic Role Labeling for Chinese verbs. In Proceed-
ings of the Nineteenth International Joint Conference
on Artificial Intelligence, Edinburgh, Scotland.
Nianwen Xue. 2006. Annotating the predicate-argument
structure of Chinese nominalizations. In Proceedings
of the fifth international conference on Language Re-
sources and Evaluation, Genoa, Italy.
438
