A Robust Cross-Style Bilingual Sentences Alignment Model
Tz-Liang Kueng Keh-Yih Su
Behavior Design Corporation
2F, No.5, Industry E. Rd. IV,
Science-Based Industrial Park,
Hsinchu, Taiwan 30077, R.O.C.
{cavs, kysu}@bdc.com.tw
Abstract
Most current sentence alignment approaches adopt
sentence length and cognate as the alignment features;
and they are mostly trained and tested in the docu-
ments with the same style. Since the length distribu-
tion, alignment-type distribution (used by length-based
approaches) and cognate frequency vary significantly
across texts with different styles, the length-based ap-
proaches fail to achieve similar performance when tested
incorpora ofdifferent styles. The experiments show that
the performance in F-measure could drop from 98.2%
to 85.6% when a length-based approach is trained by a
technical manual and then tested on a general magazine.
Sincealargepercentageofcontentwordsinthesource
text would be translated into the corresponding trans-
lation duals to preserve the meaning in the target text,
transfer lexicons are usually regarded as more reliable
cues for aligning sentences when the alignment task is
performed by human. To enhance the robustness, a
robust statistical model based on both transfer lexicons
and sentence lengths are proposed in this paper. Af-
ter integrating the transfer lexicons into the model, a
60% F-measure error reduction (from 14.4% to 5.8%) is
observed.
1 Introduction
Since the bilingual corpus is a valuable resource for
training statistical language models [Dagon, 91; Su
et al., 95; Su and Chang, 99] and sentence align-
ment is the first step for most such tasks, many
alignment approaches have been proposed in the
literature [Brown, 91; Gale and Church, 93; Wu,
94; Vogel et al., 96; Och and Ney, 2000]. Most of
those reported approaches use the sentence length
as the main feature to perform the alignment task.
For example, Brown et al. (91) used the feature
of number-of-words for alignment, and [Gale and
Church,93] claimed that better performance can be
achieved (5.8% error rate for English-French cor-
pus) if the number-of-characters is adopted instead.
As cognates are reliable cues for language pairs de-
rived from the same family, Church (93) also at-
tacked this problem by considering cognates addi-
tionally. Because most of those reported work are
performed on those Indo-European language-pairs,
for testing the performance on non-Indo-European
languages, Wu (94) had tried both length and cog-
nate features on the Hong Kong Hansard English-
Chinese corpus, and 7.9% error rate has been re-
ported. Besides, sentence alignment can also be
indirectly achieved via more complicated word cor-
responding models [Brown et al., 93; Vogel et al.,
96; Och and Ney, 2000]. Since those word corre-
sponding models, which also achieve similar per-
formance, are more complicated and run relatively
slow, they seems to be over-killed for the task of
aligning sentences and will not be discussed in this
paper.
Although length-based approaches above men-
tioned are simple and can achieve good perfor-
mance, they are usually trained and tested in
the text with the same style. Therefore, they
are style-dependent approaches. Since performing
supervised-training for each style is not feasible in
many applications, it would be interesting to know
whether those length-based approaches can still
achieve the similar performance if they are tested in
the text with different styles other than the train-
ing corpora. An experiment was thus conducted
to train the parameters with a machinery technical
manual; the performance is then tested on a general
magazine (for introducing Taiwan to foreign visi-
tors). It shows that the testing set performance of
the length-based model (with cognates considered)
would drop from 98.2% (tested in the same tech-
nical domain) to 85.6% (tested in the new general
magazine) in F-measure. After investigating those
errors, it has been found that the length distribu-
tion and alignment-type distribution (used by those
length-based approaches) vary significantly across
the texts of different styles (as would be shown in
Tables 5.2 and 5.3), and the cognate-frequency1
drops greatly from the technical manual to a gen-
eral magazine in non-Indo-European languages (as
would be shown in Table 5.3).
On the other hand, sentence length is seldom
used by a human to align bilingual sentences. They
usually do not align bilingual sentences by counting
the number of characters (or words) in the sentence
pairs. Instead, since a large percentage of content
words in the source text would be translated into
their translation-duals to preserve the meaning in
the target text, transfer-lexicons are usually used
for aligning sentences when the alignment task is
performed by human. To enhance the robustness
across different styles, transfer-lexicons are thus in-
tegrated into the traditional sentence-length based
model in the proposed robust statistical model de-
scribed below. After integrating transfer-lexicons
into the model, a 60% F-measure error reduction
(from 14.4% to 5.8%) has been observed, which cor-
responds to improving the cross-style performance
from 85.6% to 94.2% in F-measure.
The details of the proposed robust model, the as-
sociated features extracted from the bilingual cor-
pora, and the probabilistic scoring function will be
given in Section 2. In Section 3, we briefly men-
tion some implementation issues. The associated
performance evaluation is given in Section 4, and
Section 5 would address error analysis and discusses
the limitation of the proposed statistical model. Fi-
nally, the concluding remarks are given in Section
6.
1Here Cognate mainly refers to those English proper
nouns (such as those company names of IBM, HP; or the
technical terms such as IEEE-1394, etc.) that appear in the
Chinese text. As they are most likely to be directly copied
from the English sentence into the corresponding Chinese
one, they are reliable cues.
2 Statistical Sentence Alignment Model
Since an English-Chinese bilingual corpus will be
adopted in our experiments, we will denote the
source text with m sentences as ESm1 , and its
corresponding target text, with n sentences, as
CSn1. Let Mi = {typei,1,···,typei,Ni} denote
the i-th possible alignment-candidate, consisting of
Ni Alignment-Passages of typei,j, j = 1,···,Ni;
where typei,j is the matching type (e.g., 1−1, 0−1,
1−0, etc.) of the j-th Alignment-Passage in the i-th
alignment-candidate, and Ni denotes the number of
the total Alignment-Passages in the i-th alignment-
candidate. Then the statistical alignment model is
to find the Bayesian estimate M∗ among all pos-
sible alignment candidates, shown in the following
equation
M∗ = argmaxM
i
P(Mi|ESm1 ,CSn1). (2.1)
According to the Bayesian rule, the maximization
problem in (2.1) is equivalent to solving the follow-
ing maximization equation
M∗ = argmax
Mi
P(ESm1 ,CSn1 |Mi)P(Mi)
= argmax
Mi
{P(Aligned-Pairi,Nii,1 |typei,Nii,1 )P(typei,Nii,1 )}
= argmax
Mi
Niproductdisplay
j=1
{P(Aligned-Pairi,j|Aligned-Pairi,j−1i,1 ,typei,ji,1) ×
P(typei,j|typei,j−1i,1 )}, (2.2)
where Aligned-Pairi,j, j = 1,···,Ni, denotes
the j-th aligned English-Chinese bilingual sentence
groups pair in the i-th alignment candidate.
Assume that
P(Aligned-Pairi,j|Aligned-Pairi,j−1i,1 ,typei,ji,1)
≈ P(Aligned-Pairi,j|typei,j), (2.3)
and different typei,j in the i-th alignment can-
didate are statistically independent2, then the
above maximization problem can be approached by
searching for
ˆM ≡ argmax
Mi
Niproductdisplay
j=1
{P(Aligned-Pairi,j|typei,j)P(typei,j)},
(2.4)
where ˆM denotes the desired candidate.
2A more reasonable one should be the first-order Markov
model (i.e., Type-Bigram model); however, it will signifi-
cantly increase the searching time and thus is not adopted
in this paper.
2.1 Baseline Model
To make the above model feasible, Aligned-Pairi,j
should be first transformed into an appropriate
feature space. The baseline model will use both
the length of sentence [Brown et al., 91; Gale and
Church, 93] and English cognates [Wu, 94], and is
shown as follows:
argmaxM
i
Niproductdisplay
j=1
f(δc,δw|typei,j)P(δcognate)P(typei,j),
(2.5)
where δc and δw denote the normalized differences
of characters and words as explained in the follow-
ing; δc is defined to be (ltc − clsc)/radicalbiglscs2c, where
lsc and ltc are the character numbers of the aligned
bilingual portions of source text and target text,
respectively, under consideration; c denotes the
proportional constant for target-character-count
and s2c denotes the corresponding target-character-
count variance per source-character. Similarly, δw
is defined to be (ltw − wlsw)/radicalbiglsws2w, where lsw
and ltw are the word numbers of the aligned bilin-
gual portions of source text and target text, re-
spectively; w denotes the proportional constant
for target-word-count and s2w denotes the corre-
sponding target-word-count variance per source-
word. Also, the random variables δc and δw are
assumed to have bivariate normal distribution and
each possesses a standard normal distribution with
mean 0 and variance 1. Furthermore, δcognate de-
notes (Number of English cognates found in the
given Chinese sentences−Number of correspond-
ing English cognates found in the given English
sentences), and is Poisson3 distributed indepen-
dent of its associated matching-type; also assume
that δcognate is independent of other features (i.e.,
character-count and word-count).
2.2 Proposed Transfer Lexicon Model
Since transfer-lexicons are usually regarded as
more reliable cues for aligning sentences when the
alignment task is performed by human, the above
baseline model is further enhanced by adding
3Since almost all those English cognates found in the
given Chinese sentences can be found in the corresponding
English sentences, δcognate had better to be modeled as a
Poisson distribution for a rare event (rather than Normal
distribution as some papers did).
those associated transfer lexicons to it. Those
translated Chinese words, which are derived from
each English word (contained in given English
sentences) by looking up some kinds of dictionar-
ies, can be viewed as transfer-lexicons because
they are very likely to appear in the translated
Chinese sentence. However, as the distribution
of various possible translations (for each English
lexicon) found in our bilingual corpus is far more
diversified4 compared with those transfer-lexicons
obtained from the dictionary, only a small num-
ber of transfer-lexicons can be matched if the
exact-match is specified. Therefore, each Chinese-
Lexicon obtained from the dictionary is first
augmented with its associated Chinese characters,
and then the augmented transfer-lexicons set are
matched with the target Chinese sentence(s). Once
an element of the augmented transfer-lexicons set
is matched in the target Chinese sentence, it is
counted as being matched. So we compute the
Normalized-Transfer-Lexicon-Matching-Measure,
δTransfer−Lexicons which denotes [(Number of
augmented transfer-lexicons matched−Number
of augmented transfer-lexicons unmatched)/
Total Number of augmented transfer-lexicons
sets], and add it to the original model as another
additional feature.
Assume follows normal distribution and the asso-
ciated parameters are estimated from the training
set, Equation (2.5) is then replaced by
argmaxMi
Niproductdisplay
j=1
{f1(δc,δw|typei,j)P(δcognate)×
f2(δTransfer−Lexicons)×
P(typei,j)}. (2.6)
3 Implementation
The best bilingual sentence alignment in those
above models can be found by utilizing a dynamic
programming algorithm, which is similar to the dy-
namic time warping algorithm used in speech recog-
nition [Rabiner and Juang, 93]. Currently, the
4For example, the English word number are found to be
translated into ÉK,  K, ¶K, É K, É¶K, ¶
}, · · · etc., for a specific sense in the given corpus; however,
the transfer entries listed in the dictionary are ÉK and ¶
} only.
Case I (Length-Type Error)
(E1) Compared to this, modern people have relatively better nutrition and mature faster, working women marry later, and there
has been a great decrease in frequency of births, so that the number of periods in a lifetime correspondingly increases, so
it is not strange that the number of people aﬄicted with endometriosis increases greatly.
(C1) óúí, ÛHAÐôßSTXœoBEL, <vÞ¢wu, Þ>Ÿbyu×Á, øÞ2~%VãŸbó$Ó‹, äc143qc142æP66ÓÖ6.—ÑJ7SOH
(E2) The problem is not confined to women.
(E3) Sperm activity also noticeably decreases in men over forty, says Taipei Medical College urologist Chang Han-sheng.
(C2) .ÉuÞ4,ETXã4ÊûSOHuJ(, ETB¥ºc1416}péÁý'ÖEOTC»çÍ+SUBETX3L±Ç;zSOH
Case II (Length&Lexicon-Type Error)
(E1) Second, the United States as well as Japan have provided lucrative export markets for countries in this region.
(E2) The U.S. was particularly generous in the postwar years, keeping its markets open to products from Asia and giving nascent
industries in the region a chance to catch up.
(C1) wŸ, D(í1ÅêrÇ[ª¨=¹ß¹, U=¹Ë–íhEÅðDELœ}êEOTSOH
Figure 1: An illustration of length&lexical type error
maximum number of either source sentences or tar-
get sentences allowed in each alignment unit is set
to be 4 (i.e., we will not consider those matching-
types of 5−1, 5−2, 1−5, etc).
Let {s1,···,sm} and {t1,···,tn} be the paral-
lel bilingual source and target sentences, and let
S(m,n) be the maximum accumulated score be-
tween {s1,···,sm} and {t1,···,tn} under the best
alignment path. Then S(m,n) can be evaluated
recursively with the initial condition of S(0,0) = 0
in the following way:
S(m,n) = max
0≤h,k≤4
S(m−h,n−k)+score(h,k), (3.1)
where score(h,k) denotes the local scoring func-
tion to evaluate the local passage of matching type
h−k.
4 Performance Evaluation
In the experiments, a training set consisting of
7,331 pairs of bilingual sentences, and a testing
set with 1,514 pairs of bilingual sentences are ex-
tracted from the Caterpillar User Manual which
is mainly about machinery. The cross-style test-
ing set contains 274 pairs of bilingual sentences
selected from the Sinorama Magazine, which is a
general magazine (for introducing Taiwan to for-
eign visitors) with its topics covering law, politics,
education, technology, science, etc. Figure 1 is an
illustration of bilingual Sinorama Magazine texts.
For comparing the performance of alignment,
both precision rate (p) and recall rate (r), defined
as follows, are measured; however, only their asso-
ciated F-measure5 is reported for saving space.
p = [Number of correct alignment-passages in system output][Total number of all alignment-passages generated from system output],
(4.1)
r = [Number of correct alignment-passages in system output][Total number of all alignment-passages contained in benchmark corpus].
(4.2)
A Sequential-Forward-Selection (SFS) proce-
dure [Devijver, 82], based on the perfor-
mance measured from the Caterpillar User Man-
ual, is then adopted to rank different fea-
tures. Among them, the Chinese transfer lexi-
con feature (abbreviated as CTL in the table),
which only adopts Normalized-Transfer-Lexicon-
Matching-Measure and matching-type priori distri-
bution (i.e., P(typei,j)), is first selected, then CL
feature (which adopts character-length), WL fea-
ture (using word-length) and EC feature (using En-
glish cognate) follow in sequence, as reported in
Table 4.1.
The selection sequence verifies our previous sup-
position that the transfer-lexicon is a more reli-
able feature and contributes most to the aligning
task. Table 4.1 clearly shows that the proposed
robust model achieves a 60% F-measure error re-
duction (from 14.4% to 5.8%) compared with the
baseline model (i.e., improving the cross-style per-
formance from 85.6% to 94.2% in F-measure). The
5Which is defined as 2pr
p+r.
Training Set Testing Set I Testing Set II
[Caterpillar User Manual] [Caterpillar User Manual] [Sinorama Manazine]
Baseline Model 98.91 98.21 85.56
CTL 98.26 97.51 97.51
CTL+CL 99.32 98.19 89.61
CTL+CL+WL 99.61 98.83 94.07
CTL+CL+WL+EC 99.75 99.11 94.16
Table 4.1: Performance (F-measure %) of each model and SFS
result also indicates that the length-related features
are still useful, even though they are relatively un-
reliable.
5 Error Analysis
In order to understand more about the behavior
of the various features, we classify all errors which
occurs in aligning Sinorama Magazine in Table 5.1;
the error dominated by the prior distribution of
matching type is called matching-type error, the
error dominated by length feature is called length-
type error, and the error caused from both length
features and lexical-related features (either one is
not dominant) is called length&lexicon-type error6.
From Table 5.1, it is found that the matching-
type errors dominate in the baseline model. To
investigate the matching-type error, the prior dis-
tributions of matching-types under training set
[Caterpillar User Manual] and testing set II [Sino-
rama Magazine] are given in Table 5.2. The com-
parison clearly shows that the matching-type distri-
bution varies significantly across different domains,
and that explains why the baseline model (which
only considers length-based features and matching-
type distribution) fails to achieve the similar perfor-
mance in the cross-style test. However, as the 1-1
matching-type always dominates in both texts, the
matching-type distribution still provide useful in-
formation for aligning sentences when it is jointly
considered with the lexical-related feature. For
those Length-Type errors generated from the base-
line model in Table 5.1, different statistical char-
acteristics across different styles are listed in Table
6In our experiment, we do not find any error dominated
by lexical-related feature.
5.3. It also clearly shows that the associated statis-
tical characteristics of those length-based features
vary significantly across different styles. Further-
more, although English-cognates are reliable cues
for aligning bi-lingual sentences and occurs quite a
few times in the technical manual (such as company
names: IBM, HP, etc., and some special technical
terms such as RS-232, etc), they almost never oc-
cur in a general magazine such as the one that we
test. Therefore, they provide no help for aligning
corpus in such domains.
Table 5.1 also shows that errors distribute differ-
ently in the proposed robust model. The length-
type, instead of matching-type, now dominates er-
rors, which implies that the mismatching effect
resulting from different distributions of matching
types has been diluted by the transfer-lexicon fea-
ture. Furthermore, the score of erroneous lexicon-
type assignment never dominates any error found
in the proposed robust model, which verifies our
supposition that transfer-lexicons are more reliable
cues for aligning sentences.
To further investigate those remaining errors
generated from the proposed robust model, two
error examples are given in Figure 1. The first
case shows an example of Length-Type Error,
in which the short sentence (E2) is erroneously
merged with the long sentence (E1) and results in
an erroneous alignment [E1, E2 : C1] and [E3 :
C2]. (The correct alignment should be [E1 : C1]
and [E2, E3 : C2].) Generally speaking, if a short
source sentence is enclosed by two long source sen-
tences in both sides, and they are jointly trans-
lated into two long target sentences, then it is error
prone compared with other cases. The main rea-
son is that this short source sentence would contain
only a few words and thus its associated transfer-
Proposed Robust Model Baseline Mode
Error Type Percentage (%) Error Type Percentage (%)
Matching-Type Error 21.9 Matching-Type Error 81.2
Length-Type Error 62.5 Length-Type Error 14.9
Length&Lexicon-Type Error 15.6 Length&Lexicon-Type Error 3.9
Table 5.1: Error Classification while aligning Sinorama Magazine
1-0 0-1 1-1 1-2 2-1 2-2 1-3 3-1 1-4 4-1 4-2
Caterpillar 0.1% 0.25% 93.58% 2.31% 3.4% 0.05% 0.11% 0.08% 0.06% 0 0
Sinorama 0.28% 0 65.54% 1.69% 24.86% 0.28% 0 5.08% 0 1.98% 0.28%
Table 5.2: Comparison of prior distributions
Cognate Length Features (δc, δw) δcognate δTransfer−Lexicon
Occurrence Rate7 c s2c w s2w r λ µ σ2
Caterpillar 36.4% 0.65 0.87 3.45 6.09 -0.02 0.06 -0.72 0.21
Sinorama 1.1% 0.59 1.79 2.76 7.80 -0.46 0.25 -0.60 0.02
Table 5.3: List of all associated parameters
lexicons are not sufficient enough to override the
wrong preference given by the length-based feature
(which would assign similar score to both merge-
directions).
The second case shows an example of
Length&Lexicon-Type Error, in which the
source sentence (E1) is erroneously deleted and
results in an erroneous alignment [E1: Delete] and
[E2 : C1]. (The correct alignment should be [E1,
E2 : C1].) The main reason is that the meaning of
sentence (E1) is similar to that of (E2) but stated
in different words, and the translator has merged
the redundant information in his/her translation.
Therefore, the length-feature prefers to delete the
first source sentence. On the other hand, since
most of those associated transfer-lexicons in the
source sentence E1 cannot be found in the corre-
sponding target sentence C1, the Transfer-Lexicon
feature also prefers to delete the first source
sentence E1. It seems that this kind of errors
would require further knowledge from language
understanding to solve them, and is beyond the
scope of this paper.
7The occurrence rate is defined as Number of sentences
that contained congates/Total number of sentences
6 Conclusions
Although those length-based approaches are sim-
ple and can achieve good performance when they
are trained and tested in the corpora of the
same style, the performance drops significantly
when they are tested in different styles other
than that of the training corpora. (For in-
stance, the F-measure error increases from 1.8%
to 14.4% in our experiment.) The main reason
is that the statistical characteristics of those fea-
turesadoptedbythelength-basedapproaches(such
as length-distribution, alignment-type-distribution
and cognate-frequency) vary significantly from one
style to another style.
Since human align sentences mainly by examin-
ing the similarity between different meanings con-
veyed by the given bilingual sentences pair, not by
counting the number of characters in sentences, the
transfer-lexicon is expected to be the more reliable
cue than the sentence length. A robust statistical
sentences alignment model, which integrates the as-
sociated transfer-lexicons into the original length-
based model, is thus proposed in this paper. Great
improvement has been observed in our experiment,
which reduces the F-measure error generated from
the length-based model from 14.4% to 5.8%, when
the proposed approach is tested in the cross-style
case.
Last, length-features, cognate-feature and
transfer-lexicon-feature are implicitly assumed to
contribute equally in aligning sentences in this
paper; however this assumption is not usually
held because different features might have various
dynamic ranges for their scores and thus contribute
differently to discrimination power. To overcome
this problem, various features would be weighted
differently in the future.
Acknowledgement
We would like to thank both Prof. Hsin-Hsi Chen
and Prof. Kuang-Hwa Chen for their kindly pro-
viding us the aligned bi-lingual Sinorama Magazine
for conducting the above experiment. The appre-
ciation is also extended to our Translation Service
Center for providing the bilingual Caterpillar User
Manual for this study.

References

1. [Brown et al., 91] Peter F. Brown, Jennifer C. Lai,
and Robert L. Mercer, (1991). Aligning Sentences
in Parallel Corpora, Proceedings of the 29th An-
nual Meeting of the Association for Computational
Linguistics, pp. 169-176, 18-21 June 1991, UC
Berkeley, California, USA.

2. [Brown et al., 93] Peter F. Brown, Stephen A. Della
Pietra, Vincent J. Della Pietra and Robert L. Mer-
cer, (1993). The Mathematics of Statistical Ma-
chine Translation: Parameter Estimation, Com-
putational Linguistics 19: 263-311.

3. [Chen, 93] Stanley F. Chen, (1993). Aligning Sen-
tences in Bilingual Corpora Using Lexical Infor-
mation, Proceedings of the 31th Annual Meeting
of the Association for Computational Linguistics,
pp. 9-16, 22-26 June 1993, Ohio State University,
Columbus, Ohio, USA.

4. [Church, 93] Kenneth W. Church, (1993).
Char align: A Program for Aligning Parallel
Texts at the Character Level, Proceedings of the
31th Annual Meeting of the Association for Com-
putational Linguistics, pp.1-8, 22-26 June 1993,
Ohio State University, Columbus, Ohio, USA.

5. [Dagon et al., 91] Ido Dagon, Alon Itai and Ulrike
Schwall, (1991). Two Language Are More Infor-
mative Than One, Proceedings of the 29th Annual
Meeting of the Association for Computational Lin-
guistics, pp. 130-137, 18-21 June 1991, UC Berke-
ley, California, USA.

6. [Devijver 82] Pierre A. Devijver and Josef Kittler,
(1982). Pattern Recognition: A Statistical Ap-
proach, Prentice-Hall Inc., N.J., USA, 1982.

7. [Gale and Church, 93] William A. Gale and Ken-
neth W. Church, (1991). A Program for Aligning
Sentences in Bilingual Corpora, Computational
Linguistics 19:75-102.

8. [Och and Ney, 2000] Franz Josef Och and Hermann
Ney, (2000). A Comparison of Alignment Models
for Statistical Machine Translation, Proceedings
of the 38th Annual Meeting of the Association for
Computational Linguistics, pp. 1086-1090, 1-8 Oc-
tober 2000, Hong Kong.

9. [Rabiner and Juang, 93] Lawrence Rabiner and
B.H. Juang, (1993). Fundamentals of Speech
Recognition, Prentice-Hall Inc., N.J., USA, 1993.

10. [Su et al., 95] K. Y. Su, J. S. Chang and Una
Hsu, (1995). A Corpus-Based Statistics-Oriented
Two-Way Design, Proceedings of TMI-95, Vol. II,
pp. 334-353, Centre for Computational Linguistics,
Katholieke Universiteit Leuven, Leuven, Leuven,
Belgium, July 5-7, 1995.

11. [Su and Chang, 99] K. Y. Su and J. S. Chang,
(1999). A Customizable, Self-Learnable Param-
eterized MT System: Text Generation, Proceed-
ings of MT SUMMIT VII, pp. 182-188, Singapore.
(Invited Talk)

12. [Vogel et al., 96] Stephan Vogel, Hermann Ney
and Christoph Tillmann, (1996). HMM-Based
Word Alignment in Statistical Translation, Pro-
ceedings of the 34th Annual Meeting of the Associ-
ation for Computational Linguistics, pp. 836-841,
24-27 June 1996, UC Santa Cruz, California, USA.

13. [Wu, 94] Dekai Wu, (1994). Aligning a Parallel
English-Chinese Corpus Statistically with Lexical
Criteria, Proceedings of the 32th Annual Meeting
of the Association for Computational Linguistics,
pp. 80-87, 27-30 June 1994, New Mexico State
University, Las Cruces, New Mexico, USA.
