Morphological Analysis and Synthesis by 
Automated Discovery and Acquisition of 
Linguistic Rules 
Byoung- Tak ZHANG 1 and Yung- Taek KIM 
Department of Computer Engineering 
Seoul National University 
Seoul 151742, South Korea 
Abstract 
':\[his paper describes a rule-based machine learn- 
ing approach to morphological processing in the 
system called XMAS. XMAS discovers and ac- 
quires linguistic rules from examples of morpho- 
logical combinations and accomplishes the mor- 
phological analysis and synthesis by applying 
the rules. This approach is independent of lan- 
guages, saves time and effort for development 
and maintenance, and takes small lexicon space. 
A Korean version of XMAS is effectively work- 
ing in the English-Korean machine translation 
system KSHALT. 
1 Introduction 
From the knowledge engineering perspective in 
artificial intelligence, a natural language process- 
Jng (NLP) system can be seen as a knowledge- 
based system in which major concerns are the 
discovery, acqlfisition and maintenance of knowl- 
edge or rules \[2\]. 
The main purpose of this paper is to present 
a method for facilitating the development and 
maintenance of NLP systems by automating the 
discovery and acquisition of linguistic rules in 
the domain of morphology. This method was 
implemented in a morphological processor called 
XMAS as a part of the English-Korean machine 
translation project KSHALT \[4\]. 
In Section 2, we give an overview of the XMAS 
system. Section 3 illustrates the representational 
formalism. In Section 4, the processes of discov- 
ery and acquisition of morphological rules are 
described. In Section 5, the procedures for mor- 
phologicM analysis and synthesis are briefly de- 
scribed and the application results are reported. 
Section 6 concludes this paper by comparing 
XMAS with other related works and by remark- 
ing on further work. 
2 An Overview of XMAS 
XMAS (Xpert in Morphological Analysis and 
Synthesis) is a learning system \[12,13,14\] which 
consists of a learning element (Meta-XMAS), a 
knowledge base (KB), and two inference engines 
of a morphological analyzer (MOA) and a mor- 
phological synthesizer (MOS). Figure 1 shows an 
overview of XMAS and its operational environ- 
ment (parser and generator of natural language 
systems). 
I I I ~OA ~d~--~ r~.~ I 
critique 
Inductive Deductive NL System 
Learning Solving (e.g. MT system) 
Figure 1: The overview of XMAS 
1Present address: Artificial Intelligence Division, Institute for Computer Science, 
and Computational Linguistics Division, Institute for Communication Research and Phonetics (IKP), 
University of Bonn, D-5300, Bonn 1, West Germany, E-mail: ZHANG@DBNINF5.BITNET. 
i 431 
The knowledge base contains a lexicon and 
a rule base. The lexicon has entries only for 
word stems. The rule base contains the mor- 
phological rules which are learned from training 
examples. TEACHER, a human trainer, traces 
the execution of XMAS and provides training 
examples and critiques for Meta-XMAS. Meta- 
XMAS, a machine learner, acquires linguistic 
knowledge by discovering, formulating, general- 
izing and specializing grammatical rules. MOA 
solves the morphological analysis problems by 
applying the rules. MOS accomplishes the mor- 
phological synthesis by applying the rules. 
3 Description Languages 
3.1 Example Description Language 
The rule discovery procedure is initiated by 
training examples. Two kinds of training ex- 
amples are used: generalization examples and 
specialization examples. The generalization ex- 
amples are used to maintain the completeness 
of the rule base and the specializaition examples 
are used to maintain the consistency of the rule 
base \[3\]. A training example consists of a class 
name, two strings of before and after, and a cri- 
tique-an optional tag, indicating if the example 
is a specialization example. An instance of a gen- 
eralization example is (PLURAL "baby" "ba- 
bies"). 
3.2 Rule Description Language 
The formalism for the representation of morpho- 
logical rules is the string productions which are 
similar to if-then production rules \[14\]. A string 
production is a self-contained operator which 
consists of a left-hand side (LHS) and a right- 
hand side (RHS). It is symmetric and therefore 
applicable bidirectionally. 
<strlng-production> ::= ( <LHS> ------* <I~HS> ) 
<LHS> ::= (( <feature>* ) <pattern>) 
<ItHS> ::= ( <pattern> ( <feature>* ) ) 
<feature> ::= <category:> \] <conjugation:> \[ 
<prononciation> \[ <accent> \[ ... 
<pattern> ::= ( <lwindow> <main> <rwiadow>) 
<category> ::= NOUN \] VERB \[ ADJ \[ ADV I "'" 
The string production can accomodate not 
only graphemic patterns but also syntactic and 
phonological features \[1\]. A string production 
can have name(s) which is useful for direct ac- 
cess and maintenance of the rules. 
For example, the following rule is a string pro- 
duction 
(((NOUN SG) ((b) (y) (%))) --~ 
(((=) (i c s) (%))(NOUN BL))) 
where =y/ies% can be assigned to the name of 
the rule whose class name is PLURAL. 
4 Automated Learning of 
Morphological Rules 
4.1 Learning Procedure 
The procedure LEARNING (Figure 2) describes 
the top-level algorithm for rule learning in Meta- 
XMAS. 
procedure L\]~ARNING(classname, before, after~ critique) 
features ~-- LEXICON(before) 
(mop, rulename) *- DISCOVER(before, ~fier) 
H 4-- FORMULATE(cI~ssname~ ration&me, featnre6~ mop) 
matchset 4~- MA'~CH(rulename) 
if (matchset = {}) 
then CREATE(H) // a new rule H is generated /\[ 
else 
K ~ ItESOLVE(matchset) // ~elect one rule // 
if positive(crltlque) 
then GENERALIZE(It,H) // It is generalized \]/ 
else SPECIALIZE(rt,H) // It is specialized \]/ 
end LE AI'tNIN G 
Figure 2: The learning procedure of 
Meta-XMAS 
Given a training example, Meta-XMAS searches 
the lexicon for the features of the word stem 
before. By comparing the before with after, 
a micro-operator (mop) and its name is con- 
structed (DISCOVER). The mop is a descrip- 
tion of the transformation procedure from be- 
fore to after. From these, a hypothetical rule 
H is generated (FORMULATE). Then it checks 
if other rules with the same name already exist 
(MATCH). If yes, then the most special rule R 
is selected (RESOLVE) and H and R are gen- 
eralized or specialized according to the critique 
(GENERALIZE or SPECIALIZE). If no, then 
the rule H is appended to the present rule base 
with the new rule name (CREATE). 
4.2 Discovery of Rules 
The rule discovery procedure works as follows. 
First, the input strings before and after of the 
training example are compared and splitted by 
using the four pointers--/b, rb, la and ra. The 
pointers Ib and la move one grapheme at a time 
from the left end to the right, while the pointers 
rb and ra move from the right end to the left, 
until lb and la (rb and ra, respectively) point 
different graphemes. 
432 2 
This results in dividing each string into the 
three regions (lwindow, main and rwindow) 
which represent the left context, main string, 
and the right context, respectively. The cor- 
responding contexts of both strings are gener- 
alized to '=' symbol which means remains un- 
changed. The result is a micro-operator (mop). 
And then, a rule name is given to the mop by 
creating a symbol representing the transforma- 
tion form. For the training example 
(PLURAL "baby" "babies") 
the mop is 
((b) (y) (%)) --+ ((=) (i e s) (%)) 
whose name is =y/ies%. 
Finally, a hypothetical rule for the given train- 
ing ex0anple is generated by integrating the class 
name, the rule name, the features, and the mop. 
In our example, the following rule is generated 
(in a simplified form): 
=y/ies%: (PLURAL (NOUN) ((b) (y) (%)) -~ 
((=) (i e s) (%))) 
4.:~ Acquisition of Rules 
The acquisition process of rules is the iteration of 
the creation of new rules and the modification of 
existing rules. Modification is carried out by gen- 
eralization on the one hand and specialization on 
the other hand. Meta-XMAS uses the four kinds 
of induction rules in generalization \[10\]: 
1. Variablization of constant: replace a 
grapheme by =. 
2. Dropping AND conditions: decrement the 
contextual window size. 
3. Adding OR conditions: 1talon the two 
grapheme sets. 
4. Climbing generalization trees: climb the 
path to the root of the tree. 
A concept is represented as a set of graphemes 
or features which can have an explicit name. The 
generalization tree \[11\] is a tree which describes 
the generalization relations between the concepts 
in the domain. Figure 3 shows a generalization 
tree for Korean graphemes. 
AnyGrapheme 
Vowel space = Consonant* 
PVowel NVowel * Consonant 
~\]- r~_ I ~aTz. a.z~z~s~x 
Figure 3: A generalization tree for Korean 
graphemes 
Generalization using the tree is a process of 
following the path to the root on the basis of 
training examples. This is useful in concept 
learning in a well-established domain. In gen- 
eral, however, the generalization tree is not given 
a priori. In this case the Meta-XMAS builds the 
trees for itself (see Appendix A). The two gener- 
alization strategies have their strengths and \]im- 
itations. The first strategy needs more a priori 
linguistic knowledge, but it is better in learn- 
ing efficiency. The second strategy guarantees to 
find detailed rules, but is less efficient in learning 
speed. 
If Meta-XMAS applies only the generaliza- 
tion schemes, the rules learned can be over- 
generalized. So at any point (e.g. in case of 
false output) the rules should be specialized to 
avoid the overgeneralization or to handle the ex- 
ceptions. Meta-XMAS accomplishes this auto- 
matically with the aid of specialization exam- 
ples. The specialization process is a kind of 
rule creation procedure in which overgeneraliza- 
tion is checked and the overly generalized rule 
is recovered--by going down the generalization 
tree to the leaves or eliminating a grapheme from 
a concept, if necessary. 
5 Morphological Analysis and 
Synthesis 
XMAS has two inference engines (or rule- 
interpreters) of MOA and MOS. MOA, the 
morphological analyzer, solves the morpholog- 
ical analysis problms by applying the rules in 
the RHS-driven forward chaining manner \[14\]. 
MOS, the morphological synthesizer, solves the 
morphologicM synthesis problems by applying 
• the rules in the LHS-driven forward chaining 
3 433 
manner (see Appendix B). More than one mor- 
phemes can be synthesized and analyzed by an 
instruction. If more than one rule is applicable in 
rule selection, then MOA/MOS selects the most 
special rule. So the rules in the rule base need 
not be ordered, making the maintenance of the 
rule base very easy. 
XMAS is implemented on a SUN workstation 
in a Common-LISP program. A variety of exper- 
iments were carried out with XMAS on deriva~ 
tional and inflectional phenomena in almost all of 
the Korean language and part of the English and 
the German languages \[17\]. The results were suc- 
cessful in the sense that Meta-XMAS learns the 
gralnmatical rules and MOA/MOS solves cor- 
rectly the morphological analysis and synthesis 
problems by applying the rules. Part of the rules 
learned by XMAS are shown in Appendix C. A 
more efficient Korean version of XMAS, called 
MASK \[19\], was applied to the generation of 
Korean language in the English-Korean machine 
translation system KSHALT \[4\] and MASK is 
effectively working now. 
improved by the capabilities (1), (2), (4)and (6). 
The advantage (8) comes especially from (6) and 
(7). The property (9)is an indirect demonstra- 
tion of XMAS approach as a practical approach. 
As for all rule-based systems, XMAS is less 
efficient in run time in comparison with proce- 
dural systems \[6\]. There is a way to improve 
the run-time efficiency. One can construct a rule 
compiler as the transducer in KIMMO systems 
\[5,7\]. 
But XMAS is better in effectiveness in devel- 
opment and maintenance. In addition to the 
run-time efficiency, these factors should also be 
considered because, as was mentioned at the be- 
ginning, a natltral language processing system 
is a complex knowledge-based system which reo 
quires a long period of design, implementation, 
test, debugging, and extention time. 
In addition to being a practical knowledge ac- 
quisition tool, Meta-XMAS in isolation can also 
be used as a linguists' tool for scientific discovery 
\[8\] which aids linguists in the discovery and test 
of grammatical rules in morphology. 
6 Related Work and Conclud- 
ing Remarks 
In stunmary, XMAS, including Meta-XMAS as 
a linguistic knowledge acquisition tool, is char- 
acterized by the following capabilities and prop- 
erties: 
1. using graphemic, phonological, and syntac- 
tic information 
2. analysis as well as synthesis 
3. :language independent 
4. capable of treating infix inflections as well 
as prefix and suffix 
5. simple and small lexicon 
6. rules need not be orderd 
7. automatic discovery and acquisition of lin- 
guistic knowledge by inductive machine 
learning 
8. easily maintainable 
9. working in a machine translation system. 
The properties (2), (3), and (4) are the same 
as in Kaplan et al. \[5\], but XMAS has in ad- 
dition the properties (5)-(9). XMAS shares the 
properties (3) and (7) with Wothke \[15\], but is 
Acknowledgements 
This work was supported in part by IBM Ko- 
rea, Korea Science Foundation, and Department 
of Science and Technology of Korean Govern- 
ment. We would like to thank Hyuck-Cheol 
Kwon, Young-Hoon Seo, Duck-Ho Yun, Kwang- 
Seob Shim, Seung-Sik Kang, Se-Jung Kim, and 
the other members of the project KSI:\[ALT for 
their discussions and useful comments. The first 
author thanks also Prof. Dr. Winfried Lenders 
and Hooshang Mehrjerdian for careful reading 
and suggestions. 
Appendix 
A Learning 
I ,F, AILNINCt 
\[IN$1ANCE\] "l'laini.ll cXamlde: 
\[I.F2(ICON~ lt).~c~ c, lry: ( o,'¢t ~i' (At)J)) 
\[DISCOVEKER\] TrlRslMI~ Irairtln i inll~t~: 
((( ~)} (,) ((%))) -> (((t)) ~ ..- t ~) ((%)j) 
\[FDI(MUIA'I'OR\] Focnllllaled gr~mmalic~d rule: Rh (PASTX (AIM) ((\[ i)) (") ((%))) -7. ((~} (* ,. t ") {%))) 
\[RII'MA'I'CIIER\[ Malched rule: Nil. 
(I'ASFX (Al)I (((I~) (" } ((%) .> ((,) (* , \[ a ) (%~)} 
\[IIUIJ~ flASH l Ctlrfedll fide b.~.~: 
RI: =,1~% (PARI'X (AIM) (((t)) (u) ((%)}) .> ((=) ( .... ) u) (%))) 
"lk~al ntlntb¢: of tlal=: I 
a'i;;;;"~;;S",~;i~ ..................... 
\]IN, rANCH\] Training c~l~e: 
II.F.XlCOI~I ItxiLxm entry: (~ (VI:I(II)) 
\[I)\]SCOVI~IIER\] 'l'r~i~lalcd Irainin~, instance: 
(((.-)) (") ((%))) -> (((.*d)I .... I ~) ((%))) 
\[1~$\] (\[ \[ .A "O \] F~I1 l;llcd gramm~lliC~\] nile: It.~: (PAglX (Viii,lit) (((.~1) (") ~))) .> ((:) (. .. $ " ) (%))) 
\[RB-~. A" 'C tE ~\] Malch~l rule: {I'ASI'~ (ADJ) (((t)) (,) ((%))) "> ((~) (0 ~ } ") (%))) 
\[G N A >'. | Grne al/~,tl rode: 
(I'A~I'X Vlilll Al)) ( ) " % ) "> " \[~ ' ") (% 
~d new in~tance (I)A$1'X (VEgII) (~...} . %)) "> = ~ • ~ %D 
\[ItUIJ~ BASE I Curre.nt nile har~: 
P,l': ,~,/I)% (I'A~rX (VEItI! AI)J} (((,. i)) (~) ((%))) .> if-) (o ,. ~ ~) (9~,))) 
~;iJJ~i~"i;J,-.~m'~ ...................... 
IIN.'~I'ANCHI 'l','aining ex~ple: 
ilJ£XICON\] I~xk~l entry: ('.¢ ( V I-.IO1)) 
)ISCf)V £IIE (\[ • "ran.Jatcd ttaininx imt~mcc: 
fib, ) (") ((~'))) -> (((,') ~o ,, ~ ((%))) 
I :O~K U A''(') \] Finn acd £tamma cal rld~; R3: (I'ASI'X VI If{) (((~:) (w) ((%))) ,> ((:) (o "i I ~ ) (%))) 
IRII/~II{)*ICIIEI( Mmch=d role: 
IGFd~' /A ZE L/ Created role: (I'AS;rx (VEI~II) ((('I')) u ) ((%))) -> ((e) (. "r I i ) {9~))} 
\[RUI.E BAS H (:lltlCnl r lllc h~-~; 
Total numl'<r of ml~: 2 
5 435 
B Analysis and Synthesis 
SYI~'I"HESI S 
.................................................. 
\[INSTANCE\] Synthesis problem: (VASq'X ,~-) 
\[LEXICON\] Lexicon entry: C~ (ADO.) 
\[TRANSLATOR\] Translated problem in,lance: (PASTX(ADJ) (%* ~'~ ~. %)-> ?) 
\[MOS-MATCHER/Conflict set: ((I'AS'IX (VERB ADJ) (((~ \],)) (u) ((%))) -> ((=) (o ~ } ~) (%)))) 
\[ MOS- RESOLVER \] Select ed rule: (P~s-rx (VER~ ADO (((~- ~)) (.) ((%))) -> ((=) (° -,. ~ ~) (%))) 
\[GENERATOR\] Synthesized output: *11~ + PASTX--> ~ql.~.~ 
ANALYSIS 
.................................................. 
IlNSTANCE I Analysis problem: 
\[TRANSLATOR\] Translated instance: 
\[MOA-MATCHER|Conflivt set: (((PASTX (VERB ADJ) (((~ t)) (.) ((%))) -> ((=) (o ~ t ~) (%))) (x II ¢ ~ .))) 
\[MOA-~LVER1SSelected rule(s): (((PASTX (VERB ADJ) (. (o at) ,ql~ ((ADJ)))) (((-,- ~)) ) ((%))) -> ((=) ~ t (%))) 
\[RECOGNIZER\] Analysis outpu,}: ~1t%~--> -4t~((AI)J)) + I ASTX 
C 
C.1 
Part of rules learned by XMAS 
English Plural 
=/,,;% (PLURAL (N) (((o f y w p t n m thc d k r a)) NIL ((%))) -> ((=) (s) (%))) 
:=los% (PLURAl. (N) (((o ch sh x z s)) NII. ((%))) -> ((=) (e s) (%))) 
=f/vcs% (PLURAL (N) (((a e 1)) (f) ((%))) -> ((=) (v e s) (%))) 
=fe/vcs% (PLURAL (N) (((i)) (f c) ((%))) -> ((=) (v e s) (%))) 
=y/its% (PLURAL (N) (((p d b t)) (y) ((%))) -> ((=) (i e s) (%))) 
=/x% (PLURAL (N) (((u)) NIL ((%))) -> ((=) (x) (%))) 
:=on/a% (PLURAL (N) (((n)) (o n) ((%))) -> ((=) (a) (%))) 
=i/c= (PLURAL (N) (((s)) (i) ((s))) -> ((=) (e) (=))) 
:=/e% (PLURAL (N) (((a)) NIL ((%))) -> ((=) (e) (%))) 
=us/i% (PLURAL (N) (((g 1 c)) (u s) ((%))) -> ((=) (i) (%))) 
=urn/a% (PLURAL (N) (((d t)) (u m) ((%))) -> ((=) (a) (%))) 
=/rcn% (PLURAL (N) (((d)) NIL ((%))) -> ((=) (r e n) (%))) 
=/on% (PLURAL (N) (((x)) NIL ((%))) -> ((=) (e n) (%))) 
:=We= (PLURAL (N) (((m)) (a) ((n))) -> ((=) (e) (=))) 
=oNcc= (PLURAl. (N) (((t g O) (o o) ((s t))) -> ((=) (c e) (=))) 
C.2 Past Tense of Korean 
== \]-'8/ )\]at % (PASTX (ADJ) ((( \]- .~.) (~;' ~)) ( \]- "8) ((%))) -> ((= =) ( II at) (%))) 
==1. % (PASTA (VERB) (((z 7) (}*)) NIL ((%))) -> ((= =) ( ) (%))) %=~-,- I-'~ % (PASTX (VERB) (((%) (, o )) (~) ((%))) -> ((% =) (_,_ ~. ) (%))) 
==~/~oA% (PASTX (VERB) (((~) (~)) (~) ((%))) -> ((= =)(~ o ~ .) (%))) 
==~/~ ~% (PASTX (VERB) frO' ~ ) ( "1 --)) (~) ((%))) -> ((: =) (~ o -I * ) (%))) 
==/~.% (PASTX (VERB ADJ) (((~) (__)) Nm ((%))) -> ((: =) (~ q *) (%))) ==_._J~% (PASq'X (VERB ADJ) (((_t_ \]-) (~.)) (_) ((%))) -> ((= =) (~ \]- . ) (%))) 
=:.__/~% (PASTX (VERB) (((_) (~))(_) ((%))) -> ((= =) (~ ~ at ) (%))) 
==, /~% (PASTX (VERB ADJ) (((~ Y ) ( \]- .,..)) (,) ((%))) -> ((= =) (o ~ }. at) (%))) 
---=./~% (PASTX (VERB ADJ) (((7 t. ~ o) (-r- "I 4)) (u) ((%))) -> ((= =) (o T "I ~) (%))) ==); /o~% (PASTX (VERB ADJ) (((x ~-) (}-)) (z) ((%))) -> ((- =) (o }- ~ ) (%))) 
==x ;~o/o (PASTX (VERB) (((. 7) ('1" --)) (x) ((%))) -> ((= =) (o -\] at) (%))) 
==/~% (PASTX (VERB) ((('8) (}-)) NIL ((%))) -> ((= =) ( o :\] at) (%))) 
%=-i-I "\]* % (PASTX (VERB) (((%) (~)) (T) ((%))) -> ((% =) ( { * ) (%))) %=__/ -\]at % (PASTX (VERB) (((%) Cat )) (--) ((%))) -> ((% =) ( q ~ ) (%))) 
==1~,% (PASTX (VERB ADJ) (((.x. }-) (" z ~ :7 i~ 7 )) NIL ((%))) -> ((= =) (o I" '~) (%))) 
==/~% (PA~FX (VERB) ((( -\] \] -1-) (~ "¢ ~-)) NIL ((%))) -> ((= =) (o "l "~) (%))) 
436 6 

References 

\[1\] Bear, J., A morphological recognizer with 
syntactic and phonological rules, Proc. of 
COLING-86, pp. 272-275, Bonn, 1986. 

\[2\] Hayes-Roth, F., Waterman, D.A., and 
Lenat, D.B., Building Expert Systems, 
Addison-Wesley, 1983. 

\[3\] Jansen-Winkeln, t~.M., Induktives Lernen 
yon Granunatikregeln aus ausgewaehlten 
Beispielen, In: Savory, S. (ed.), Kuenstliche 
Intelligenz und Expertensysteme, pp. 211- 
223, Oldenbourg, 1985. 

\[4\] Kang, S.S., Shim, K.S., Zhang, B.T., 
Kwon, H.C., Woo, C.S. and Kim, Y.T., An 
English-Korean system for hmnan-assisted 
language translation, Proc. of TENCON- 
87, IEEE, pp. 550-555, 1987. 

\[5\] Kaplan, R. and Karttunen, L., Computa- 
tional Morphology, CSLI Report No. LI 283, 
Jlme 1987. 

\[6\] Kim, S.Y., Choi, K.S., and Kim, K.C., A 
Korean morphological analyser using tab- 
ular parsing and connection information, 
Proc. of 87 Spring Conf, on Arti. Intell., 
Korea Info. Sci. Soc., pp. 133-147, 1987, (in 
Korean). 

\[7\] Koskenniemi, K., Two-level model for mor- 
phological analysis, Proc. of IJCAI-85, pp. 
683-685, 1985. 

\[8\] Langley, P., Zytkow, J.M., Bradshaw, G.L., 
and Simon, H.A., Three facets of scientific 
discovery, Proc. of IJCAI-83, pp. 465-468, 
1983. 

\[9\] Lenders, W. and Will4e, G., Linguistisehe 
Datenverarbeitung: Ein Lehrbueh, West- 
deutscher Verlag, 1986. 

\[10\] Michalski, R.S., A theory and methodology 
of inductive leaxning, in: Machine Learning, 
Michalski et al. (eds.), Tioga press, pp. 83- 
134, 1983. 

\[11\] Mitchell, T.M., Utgoff, P.E. and Banerji, 
R., Learning by experimentation: acquiring 
and refining problem-solving heuristics, In: 
Machine Learning, Michalski et al. (eds.), 
Tioga press, pp. 163-190, 1983. 

\[12\] Smith, R.G., Mitchell, T.M., Chestek, R.A., 
and Buchanan, B.G., A model for learning 
systems, Proc. of IJCAI-77, pp. 338-343, 
1977. 

\[13\] Veenker, G. (ed.), Automatisches Lernen, 
Arti. Intell. Seminar, Inst. for Comp. Sci., 
Univ. of Bonn, 1990. 

\[14\] Waterman, D.A. and Hayes-Roth, F., An 
overview of pattern-directed inference sys- 
tems, In: Pattern-Directed Inference Sys- 
tems, Waterman, D.A. and Hayes-Roth, F. 
(eds.), Academic press, pp. 3-22, 1978. 

\[15\] Wothke, K., Machine learning of morpho- 
logical rules by generalization and analogy, 
Proc. of COLING-86, pp. 289-293, Bonn, 
1986. 

\[16\] Wothke, K., Maschinelle Erlernung und Si-. 
mulation morphologischer A bleitungsregeln, 
Dissertation, IKP, Univ. of Bonn, 1985. 

\[17\] Zhang, B.T. and Kim, Y.T., Macldne learn- 
ing of linguistic rules for multi\]ingual mor- 
phological analysis and synthesis, Journal 
of Korea Info. Sci. Soc., in press. 

\[18\] Zhang, B.T., Yoo, S.I, and Kim, Y.T., 
XMAS: An expert in Morphological Analy- 
sis and Synthesis, Proc. of '88 Spring Conf. 
on Arti. Intell., Korea Info. Sci. Soc., pp. 
179-188, March 1988. 

\[19\] Zhang, B.T. and Kim, Y.T., Morphologi- 
cal synthesis and analysis of Korean: a ma- 
chine learning approach, Proc. of 87 Au- 
tomn Conf. of Korea Info. Sci. Soc., pp. 
573-576, Sept. 1987. 
