A Part-of-Speech-Based Alignment Algorithm 
Kuang-hua Chen and tlsin-I Isi Chen 
Department of Computer Science and Information Engineering 
National Taiwan University 
Taipei, Taiwan, R.O.C. 
e-mail: hh chen@csie.ntu.edu.tw 
Abstract 
To align bilingual texts becomes a crucial issue 
recently. Rather than using length-based or 
translation-based criterion, a part-of-speech-based 
criterion is proposed. We postulate that source texts 
and target texts should share the same concepts, ideas, 
entities, and events. Simulated annealing approach is 
used to implement this aligmnent algorithm. The 
preliminary experiments show good performance. 
Most importantly, the experimental objects are 
Chinese-English texts, which are selected from 
different language families. 
1. Introduction 
Real texts provide the alive phenomena, usages, and 
tendency of langnage in a parlictflar space and time. 
This recommends us to do the researches on the 
corpora. Recently, many rese~{rchers timber claim 
that "two languages art more informative than one" 
(Dagan, 1991). They show that two languages coukl 
disambigna.te each other (Gale e¢ al., 1992); bilingual 
corpus could form a bilingual dictionary (Brown et al., 
1988) and terminology correspondence bank (Eijk, 
1993); a refined bilingual corpus could be formed the 
examples for machine translation systems (Sumita et 
al., 1990). To do such kinds of researches, the most 
impmlant task is to align the bilingual texts. 
Many length-based alignment algorithms have 
been proposed (Brown et al., 1991; Gale and Church, 
1991a). The correct rates are good. However, the 
languages they processed belong to occidental family. 
When these algorithms are applied to other rtmning 
texts from different families, will the performance 
keep on tile same level? Other translation-based 
alignments (Kay, 199l; Chen, 1993) show the 
difficulty in determining the word correspondence and 
are very complex. 
In tiffs paper, we will introduce a part-of-speech 
(POS)-based alignment algorithm. Section 2 will 
touch on the level of alignment and define the 
sentence terminators. In Section 3, we will propose 
tile criterion of critical POSes and investigate the 
distribution of these POSes in the Chinese-English 
texts. Section ,l will describe a fifir and rigorous 
method for evaluating performance. Then, we apply 
simulated annealing technique to conducting 
experiments and show tile experimental results in 
Section 5. Section 6 will give a brief conclusion. 
2. Alignment Problem 
Alignment has three levels: 1) paragrapll; 2) sentence; 
and 3) word. Paragraph level is sometimes called 
discourse level. Many efforts are involved in senlence 
level and fewer researchers louch on the word level 
(Gale and Clmrch, 1991b). To do sentence alignment, 
we should first define what a sentence is. ILl English, 
tile sentence terminators are fifll stop, question mark, 
and exclamation mark. \[\[owever, tim usage of 
punctuation marks is unrestricted in Chinese and the 
types of punctuation marks are numerous (Yang, 
1981). Nevertheless, in order to parallel the languages, 
we define that tile sentence markers are fldl slop, 
question mark, and exclamation mark over all 
languages. Therelbre, an alignment of two texts is to 
find a best sequence of sentcnce groups, which arc 
ended with one of tile sentence ternlillalors. 
Following Brown el al. (1991), we use tile term 
bead. A bead contains some sentences of source and 
largct texls. Thus, alignment can be defined as (1). 
(I) An alignment is to find a bead sequence under 
some crileria. 
If Ihe applied criteria are significant, the performance 
will be good. Finding significant criteria is tile core of 
this research. 
3. Criteria of Alignment 
ALLy aligmnenl algorilhm has its own criteria. For 
example, many alignment algorithms are based on 
sentence lengfl~ and word correspondence. Here, wc 
propose a POSes-based crilerion. 
(2) Alignment Crilerion: 
"File numbers of critical part of speeches (POSes) 
of a langt,age pair in an aligned bcad are close. 
166 
Now, the problem ix what forms the critical POSes. 
Following many gr~,mmar formalisms (Sells, 1985), 
the content words will be the good indicalors. 
Therefore, we think nouns, verbs, and adjectives as the 
critical POSes. In addition, we inch,de mm~bers and 
quotation marks in the critical POSts due to intuition. 
The English tagging system used in this work follows 
that of the LOB Corpus (Johansson, 1986). The 
Chinese tagging system follows that of the BDC 
Corpus (BDC, 1992) but with some modilications. 
The BI)C COrlms docs not assign tags to Imnctualion 
marks. We adopl Ihe same philosophy of I,OB Corpus 
to assign the tags of the punctuation marks as 
themselves. These critical POSes in English and in 
Chinese arc listed in Table 1. N- represents all lags 
initialed with N, i.e., - is a wildcard. 
Notln 
Verb 
l- Number 
\[ Quotation Marks 
Tahle I. Critical POSes 
\]\[ I;,,mish (t 01+ t+, ,s) C'hi,,csc +I 
I A+-7--- U---- 
Our bilingual corpus is investigated to check the 
effectiveness of the Imstulation (2). Ten aligned 
Chinese-to-English texts, CE 01 to CIQI0, are 
considered as the objects of experimenls. These texts 
are selected front Sinorama Magazine, published in 
Chinese and English monthly by Gove,mnenl 
Information ()ffice of R.O.C. Appendix lists the 
source of these ten texts. We compute tile average of 
differences (AD), wlriancc of differences (VI)), and 
standard deviation of differences (SD) of the critical 
POSes. *l':tblc 2 itemizes tile wdues. 
'l'able 2. Statistics of Bilingual Texts 
Files CE 01 ~ (;1'; 02 CE (13 I CE 0,1 I CI". 05 CE 06 CE 07 CE 0~ CI,; 09 (21,; 10 Total 
39 I Z 23 _ ~__5:+ _2' 37 .',55 / 
AI) 1.857 2.103 2.069 2.758 I 2.192 2.043 2.3602.509 2_.759 2.541 2.379 
5"093~ A-I 7.078 ~2.g2,~ 5.990 5.9R6 5.21g_g 3.438 4.8111_ No.. vg~ 2.c,94 4.3,1~ 3.375 ...... 
____ SI) 1.641 2.085 1.83~ 2.2 S I 2.661 1.681 2.,1,18 2.,I'177 2.284 . 1.85,1 2.19:~_ 
AI) 1.000 1.153 1.724 2.333 I 1.615 1.522 1.680 2.189 1.776 1.757 1.758 
Ve,-t, Vl~_) 1.429 ,1.0,1~ 2.613 a,!l,tJ I 453_ 1~72~. 1 gg~_ 226(, _ 2.2T/ _!+96~ 2=53!_._ 
--SI) 1..195 2.O11 1.(,17 I.S~:~l 1.361 1.2~1,1 1.1JR 1:50~ I.)07._ I.,103 _. 1.5') I ~ 
AI) 11.929 \[ I.,162 2.310 1 061 I 1.308 2.391 0.8R0 1.679 2.379 1.459 1.64'I 
Adj. VI)) ().781___ 19412.69_7_ ().g'15 \[ 1.,I,1.I 3.282 _ 0.586 2.067 3~6R,I 2.032 2.372 
-- S I).. 0.894 1.391~ 1 6,12 0919 I 1.2112 1.812 0.'/65 1.43R 1.91') IA26 1.5'111 .... 4 ...................... 
AD 0.321 0.949 1/.655 (1,636 I 11,769 1.217 11.600 //.736 0.500 0,838 0.7117 ........ ----4 ......... 
Nt.n. VD 0.290 I.690 0.502 0777 1_ 0.716 _ 1.47,I ().72(1 0.949 ().73~3 i~.d_1.055 0.948 
At) 0.357 0,282 0.345 0.3031 0.923 0.609 0.520 0.4RI (1.207 0.432 0.416 
euot. VI) 0.5~7 0.351, 0.77R ~151,1 I 1.'_)1.7 1.1_9 s .... 2.{)H)) .... 1+193 (.}.509 i 0.840_. 0.9,19 ~ 
__ ,RI)_) 0.766 0.597 (I.882 ()) 71"7~ 1.3~5 1.093 I'10~_~ _1092 _(2.71:C 0.917 Al.,;'/,!q / 
Tolal Vl) 3.667 11+173 ~ 12.61,1 ,19S'~ I 8534 7.7(/9 10.790 I,1.607 1(1.559 5.182 9.71,1 / 
__ m~ 9.9ts_ ___A.226 L_?'22L 2~?vd, 3.2.~2 3.~J _ 3.249 L 2.276 3.1vt 
167 
For oriental languages like Chinese, the 
correspondence with its aligned counterpart in 
occidental languages is not so manifest like alignment 
of two languages within the same family. On the one 
hand, the POSes may be changed within an alignment; 
on the other hand, a sequence of English words may 
correspond to a Chinese word. These phenomena 
make the alignment much harder. HowEver, Table 2 
shows that these POSes are good indicators for 
alignment. 
4. How to Evaluate the Performance 
Before proceeding the experiment, another important 
issue is how to ewfluate the correct rate for an 
alignment. No literatures tonch on this issue. We just 
find in literatures what performance is rather than how 
to evaluate it. Note that aligmnent has the order 
constraint. On the OnE hand, when an error occurs, 
the performance should drop quickly. On the other 
band, the error will not broadcast to the next 
paragraph. That is to say, the error will be limitcd in a 
range. Our criterion for evaluating performance takes 
care of the two factors. 
For a given text, we could manually find the real 
alignment. This alignment consists of a sequence of 
beads, as mentioned previously. We call the sequence 
of beads Real Bead Sequence (RI¢S). In contrast, we 
may apply any alignment algorithm to finding an 
alignment. We call this aligmnent Contputed Head 
Sequence (CIL'¢). 
Ill order to evahlate the perforlnancc of tile 
alignment algoritlun, WE filrthcr define the 
lncrentental Bead A'eqHence (II}S). 
(3) hTcremental Bead Sequence 1BS of a given 
bead sequencc BS is a bead scquence, such 
that \[lead \]fli in IBN is snmmation of 15 
(0 _<. j .<_ i - 1) in IJS. 
Therefore, two possible II3Ses, 1RBS and /C\]IS, are 
generated under this consideration. We define 
performance for an alignment as 
(4) Performance = 
nutnber of common beads in IRI3S and ICBS 
number of beads in IRBS 
Table 3 demonstrates how to calculate lhc 
performancE. Two beads, (3,3) and (10,10), are 
shared by II?BS and ICI}S. The total numllcr of beads 
in IReS is 10. Therefore, the performance is 20%. In 
the following experiment, we will use this method to 
evahmte performance. 
Tahle 3. ExamplEs for Calculating Perfimnance 
RBS 
I R BS 
CBS 
ICeS 
(I,0), (1,1), (l,2), (1,1), (I, 2), (1,1), (1,1), (I,0), (1,1), (1,1) 
(1,0), (2,1), (3,31, (4,4), (5,6), (6,71, (7,81, (g,8), (9,9), (10,10) 
(1,1), (1,1), (I,1), (1,0), (1,1), (1,1), (1,2), (2,1), (I,2) 
(1,l), (2,2), (3,3), (4,3), (5,4), (6,5), (7,7), (9,81, (10,10) 
__p~formance if_ 2/1/} = 0.2 = 2/}% 
5. Alignment Algorithm 
The alignment algorithms proposed in the past 
literatures try to find an optimal alignment which Ires 
the largest alignment probability. Due to llle very 
large search space, they all consider only five types of 
beads: (0,1), (1,1), (1,2), (2,11, and (1,0). After 
examining our corpus, we can find other types of 
beads such as (1,3) and (1,4). Furthermorc, bcad type 
(2,4) is also found. Table 4 lists tile distribution of 
bead types in the testing lexts. Eiphl bead types appear 
in tile bilingual texls. Bead type (l,l) is the majority 
(63.9%). Bead types (1,31, 0,4), and (2,4), which :ire 
not treated in other papers, occupy 8.2%. If the 
alignment algorithm did not dcal with these bead types, 
the correct rate would be bound to 91.8%. 
Tahle 4. Distrihution of Bead Types 
 eadType\[I (1.0) <0.1) (1.,)\[(2.,) (,.2) 1('.3) (,.4) <2.4) I'l ota, 
%, 0.56 0.56 63.94 I 1.6,) 25.07 .... 5.63 2.25 0.28 l{}O 
168 
It shows tile difficulty of tile alignment task. If we 
allow various types of beads and adopt the optimal 
search, tile processing cost is too high to stand. A 
good algorithm should satisfy tile following two 
conditions: 
It is a general local search algorithm. 
It allows the unlinfited bead types in the 
aligning process. 
Under this consideration, simulated annealing 
approach (Aarts and Korst, 1989) is used to align texts. 
The idea of annealing comes from condensed matte," 
physics. It involves two steps: 1) increasing 
temperature of matter; 2) decreasing temperature 
gradually until the matter in the ground configuration. 
Simulated annealing is to sinudate the almealing 
process. Therefore, a simnlated annealing mechanistn 
is composed of four parts: configuration, (ransition 
fimction, energy fimction, and annealing schedule. If 
we take an alignment as a configuration, the possible 
alignmenls constitute tile configuration space. In 
addition, every configuration is associated with an 
energy. The optimal configuration is tile one which 
has tile lowest energy. Simulated annealing is to find 
tile optimal configuratio,i from an initial configuration 
by generating a sequence of configurations under a 
control parameter. 
For our application, we introduce another 
component, Transition Vector. The five components 
are defined as follows. 
(5) Configuration (C): An alignment is a 
configuration naturally. For example, a 
possible bead sequence, {(1,21, (I,1), (1,1), 
(1,2), (1,1)}, is a configuration. 
(6) 7)'ansition l,'unction (T): Given a 
configuration, this fimction is responsible for 
generating its next configuration. A 
transition vector is generated ill random, alld 
then tile transition ftmction moves one 
configuration to another configuration 
according to the transition vector. 
(7) 7}'ansition I/ector (TV): A transition vector 
consists of 4 components (H, N, IV, D). 
B denotes the identification (counted from 01 
of a selected bead. 
N specifies whether to generate a new bead 
or not. IfN equals to 0, no new bead ix 
generated. If N equals to 1, a new bead is 
generated. 
IV represents which language ill tile selected 
bcad should be moved ont. If W equals to 0, 
one of the marginal sentences of tile first 
language should bc moved out. Otherwise, 
one of tile marginal sentences of the second 
language should be moved out. 
I) represents the moving direction. 0 denotes 
the left marginal sentence of the selected 
bead is moved left, and 1 denotes the right 
marginal sentcnce of the selected bead is 
moved right. 
For example, transition fimclion will transit 
a configuralion {(1,2), (1,1), (1,1), (1,21, 
(1,1)} to {(1,2), (l,l), tO, l), (1,0), (1,2), 
(1,1)) according to tile transition vector TV 
-(2, 1,(1, l). 
(s) lOwrev I,)mclion (E): Assume each sentence 
has a weight, which is measured by tile 
nmnber of critical POSes. The weight 
difference of a bead is the difference between 
the weighls of respective sentences in one 
bead. The energy of a configuration is the 
sum of weight differences of all beads in a 
configuration. 
(9) Annealing Sclwdule (AS): When a new 
conti~,uu'ation (" is generated, two 
alternatives are considered: move to the new 
configuration C' or retain tile current 
configuration C'. 'File criterion is if K(C') <" 
E(C), the new configuration is adop|ed. 
1 lowever, if 
exp (1¢(C) -. 1,2(C' )) > random 10,1 
c:pk 
we ,,viii also move Io the now confil,tlratmn. 
()thep, vise, the cllrrctlt configtlralioll is 
retained. This is the well-known Metropolis 
('rileri(m. The CPk is Ihc control parameter, 
which will be reduced gradually in tile 
a,mealing process. 
Now, we apply tile simulated annealing to aligning 
the texts, CE 01 to CE 10. Tile initial control 
parameter cpk is 1.0 and initial nm length L k is I000. 
Wc reduce the control parameter with 0.5% after each 
rim. Tile initial configuration is randomly generated. 
Wc conduct two cxpcriments, 1) without using 
paragraph markers; 2) with using paragraph markers. 
'Fhc results arc shown in Table 5 and Table 6, 
respectively. 
169 
Table 5. Correct Rate for Simulated Annealing (without using paragraph marlter) 
creXt s orrect 
Total 
Correct P~tte 
33 27 \] 23. 
'0.788 0.778 I 0.870 0.880~}6 1"0.803 0.730 0.789 1 0.718 0.793 
Table 6. Correct Rate fl)r Simulated Annealing (with using i)ar'4graldl marker) 
Texts " lIcl,:_of I c~,:.o2-\] c1~,_o3 \] el.; 04 \] CF...o.s-I-c,,:_o~c,': o9 I c!,:.lo \] Total 
Correct 28 36 28 30 26 23 25 49 55 35 335 
Total 28 39 29 33 27 23 25 53 61 37 355 
Correct P~tte 1.000 0.923 0.966 0.909 0.963 1.000 1.000 0.925 0.902 0.946 (I.944 
Correct 
Rale 
1 
0,9 
0.11 
03 
0.6 
0.5 
0.4 
0.3 
0,2 
0.'t 
O i 
0 
\[ ~ ,qilmdaled Almealing without p~lragraph Iilatkel" ..... • ~ ..... Simulated Annealillg willl p~ragraph lllarker 
: : .... I I I 
2 4 6 8 10 
Texts 
Figure 1. Comparison of Aligmnent PErformance 
The correct rates without and with using paragraph 
markers are 78.9% and 94.4%, respectively. The 
latter result (94.4%) is better than the botmd eorrecl 
rate (91.8%) mentioned before. It shows that those 
difficult bead types are resolved in our approach. 
Comparing Tables 5 and 6, we conclude Ihat when the 
paragraph markers are used, the performance 
increases significantly. Fig. 1 shows the significance 
of paragraph markers. In other words, if an alignmenl 
algorithm could use any reliable anchor points in IhE 
texts, the performance will incrEasE sharply. 
In fact, the pErlbrmancc of alignmcnt is depcndenl 
on the naturc of the texts. When aligning a noisy texts 
without rcliablc anchor points, we will definitely do a 
bad job. However, the simulated annealing approach 
could reduce the risk, and the performance will keep 
over 78% in our experiment. 
6. Conclusion 
A new criterion to aligning texts is proposed in this 
paper. The criterion is based on an observation that 
the source texts and thc target texts should share the 
same concepts, entities, ideas, and events. Sentence 
length (no matter word-based or character-based) 
(Brown, el al., 1991; Gale and Church, 1991a), is not 
so critical on languages across different language 
families. Translation-based crilcrion (Kay, 1991; 
ChErt, 1993) is very. uscfifl, but il is also very complex. 
Surely, to decide word correspondences is difficull. 
Our criterion provides a tradcoff between the length- 
based criterion and the translation-based criterion. 
The elucs of critical POSes are parlially syntactic and 
partially statislic~d. 
ThE performance of simulated annealing approach 
to alignment is 94% in our Experiment, if we use the 
paragraph markers. Without paragraph marker, the 
value drops lo 78%. GEnerally speaking, it works well 
for languages across different language families. 
The main conlribntion of lhis work is to provide an 
alignment algorithm for aligning oriental langvages 
with occidental languages. The fitture work should 
focus on the large experiment, normalizing the weight 
of critical POSes and other search techniques. 
170 
Acknowledgements 
Research on this paper was partially supported by 
National Science Council grant NSCg,l-(}40g-l~002- 
005. 
References 
Aarts, E. and J. Korst (1989). Smndated Annealing 
and Bollzmann \]v\[achines, John Wiley & Sons. 
13ehavior Design Corporation (1992). 7he BDC 
Chinese Tagged Corpus, Taiwan, R.O.C. 
Brown, P. el al. (1989). "A Statistical Approach to 
Language Translation." Proceedings of COLING, 
pp. 71-76. 
Brown, P. et al. (1991). "Aligning Sentences in 
Parallel Corpora." Proceedings of 291h Annual 
Meeting *f the ACL, pp. 169-176. 
Chcn, S, (1993). "Aligning Sentences in Bilingual 
Corpora Using Lexical Information." 
Proceedings of 3 Ist Annual ~\[eeting of the A (?L, 
pp. 9-16. 
Dagan, l., A. Imi and.lJ. Schwall (1991). "Two 
Langtlages are More Informative Than One." 
Proceedings of 29th A,nual Meeting of flu, A (7, 
pp. 130-137. 
Eijk, P. (1993). "Automating the Acquisition of 
Bilingual Ternlinology." l'roceedinvs of the 
Sixth Conference of the l?uropean Chapter of the 
ACL, pp. 113-119. 
Gale, W. and K. Church (1991a). "hlenti~,ing Word 
Correspondences in Parallel Texts." 1"roceedings 
of lrburth DARPA Work,vhop on Xpeech and 
Natural Language, pp, 152-157. 
Gale, W. and K. Church (1991b). "A Program for 
Aligning Sentences in 13ilingual Corpora." 
l'roceedings of 29th tlnnual A,/eeting of the d CI,, 
pp. 177-184. 
Gale, W., K. Church and S. Yarowsky (1992). "Using 
Bilingual Malerials to Develop Word Scnse 
Disambiguation Methods." Proceedings of the 
leottrth International Conference on 77ieorelical 
and Methodological Issues in J:\[achine 
7)'anshttion, pp. 101-112. 
Johansson, S. (1986). 77re Tagged LOll ('OrlmS: Users' 
Manual, Bergen: Norwegian Conipuling Centre 
for lhe I lumanilies. 
Kay, M. (1991). "Text-Translation Alignnlcnt." 
Cor!ference Handbook of A CI\[hlLLfYgl: A4aking 
Connections, Tempe, Arizona, p. 257. 
Sells, P. (1985). Lectures on Contemporary ,%ivntactic 
Theories, Lecture Notes, No. 3, CSLI. 
Sumita, E., H. lida and H. Kohyama (1990). 
"Experiments and Prospects of Example-Based 
Machine Translation." Proceedings of the Third 
International Conference on 771eoretical and 
Methodological Issues in J4achine 7Yanslation, 
pp. 2(13-212. 
Yang, Y. (1981). l?esearches on l)unctnation g&.'ks, 
Tien-Chien f'ublishing Company, Hong Kong. 
Appendix 
The testing corpus, CE__01 to CE_I0, are selected 
from Sinoranm Magazine O~ II!~l'f::ii,~0. The details of 
these texts are lisled in the following. 
CI!_01' ~lJ~'1~yj (Lin, Yung-fang/tr. by Phil Newell), 
.)~1, '-I I I " ~ ¢<~" ''¢" '/ ,'-'~<,~t- J J-J, ,~'~,k)~)J't ~J (Is This Any Way to 
Ibm a School '? )," )/~ ~)~;.~(:,/E(~ (Sinorama 
A&~gazine), Jan. 199l, pp. 108-111. 
CE02:j~J~{514 (Chao, Shu-hsia/tr. by Peter Eberly), 
" tiff; iliL ~ ~.! (A Member of the Overseas 
Chinese Community )," )\[j//!.~ff\[)/~: (Sinoranm 
A4agazine), Mar. 1991, pp. 110-111. 
CE_03: 9)~I!VJ ' ?Ai (Chang, Chin-ju/t,'. by Phil Newcll), 
.. r ~,I~ )k _l I'(g ~\['i '\]'~/ -- ~; J~z (It's Hard to 
Conceivc--Infertility in Taiwan)," )/';//!~'/~(/,{~: 
(>,'inoranla A,h~gazine), May 1991, pp. 22-23. 
t ,i I: CE_t)4: Ji,~)fi (Ch'i Chiin/tr. by Peter Ebcrly), h\]l 
(IdLe Thoughts)," )/," r}!: .~/~ ~,~. (Sim~rama 
Magazine), May 1991, pp. 94-95. 
CE_05: \]~f\[!~ (Chcn, Elaine/tr. by Peter Eberly), " 
)JH ')'H-fill \[: fl<3 r ~,i~ ~/~. I :l- '~I~ (Caltcch's 
"Cxack-Troop" Way of Life)," )/j ~)!~,ff/~,/~: 
(Ninoramo Magazine), June 1991, pp. 124- 
125. 
CE_06: -~,<-)~. (l_,i, Laura/tr. by Christopher I lughes), 
. ,.i(~l,~.~' 1- r ~f(i,~,,,l'l~\]~!~ A (A Curse on P;oth 
Our llouses)," )\[~; //!: ~,'~/~ ,;~(: (Sinorama 
MagazimO, Sept. 1991, pp. 40-41. 
CE07: {t~ ~,,c (Wci, Ihmg-chin/tr. by Christopher 
l lughes), " r ,,(}~ i~L _1 ~,!-(f:,~fl:-f J ::~ (Cholera- 
Present Progressive Tense)," )/~ ~: if{; ,~: 
(Sinorama Magazine), Nov. 1991, p. 47. 
CE08: "~g~y},~ ~"i-(Wci, lhmg-chin/lr, by Phil Newcll), 
"/\]x,(~, Y ': ~ <s ':='~-:~ " ~' ' 
Vinlses - II Can l lappen to Yon)," 3\[~//!:ff\[,jt~: 
(>,'inoranm Magazine), April 1992, pp. 34-38. 
CE_09: \]t~,li}~}: (Ten\[,, Sue-feng), " ,17~:;i,~lil~l,i\[',(~,l-- 
~- ~'fl~Sf' '1' 1'10 ",',{~)H,~<f!~/'/~ I,'~I~;~ ~r " ' 
(Rediscovering Asia The International 
Confercnce on "The Asian Regional 
Econonly")," )\[~ "l!-~/~;/*~: (Ninorama AiIogaz# e), 
June 1992, pp. 22-26. 
CE_I0: ~@ i,\]'j~ .~,- "" (l,in, Ching-ynn/tr. I)7. Jonathan 
Barnard), " t ~,: '-,~,i,~ ~. ,, I--~,~1 ,-~LG (Book Review -- 
Mourning My Breast)," )\[~ ,'),~;~,~/~;~: (Sinorama 
A4agazine) Feb. 1993, pp. 90-92. 
171 
