A Phonological 
Abstract 
This paper describes the framework of a Korean 
phonological knowledge base system using the unification- 
based grammar formalism : Korean Phonology Structure 
Grammar (KPSG). The approach of KPSG provides an 
explicit development model for constructing a computational 
phonological system : speech recognition and synthesis system. 
We show that the proposed approach is more describable 
than other approaches such as those employing a traditional 
generative phonological approach. 
1, Introduction 
Knowledge Base System Using Unification-based Formalism 
A Case Study of Korean Phonology 
Hee-Sung Chung 
Korea Academy of Industrial Technology(KAITECH) 70-6, Yangjae-Dong, Seocho-Gu, S6oul Korea 
syllable consists of all the segments that precede the neeleus 
and are tantosyllabic segments that follow the necleus. A 
syllable that has no coda is called "open syllable" : one with 
a coda is called "closed syllable". According to the above 
definitions of a syllable structure, the Korean syllable can 
be described as follows : 
This paper has two main goals: one is to investigate 
the phonological "knowledge base system for constructing a 
Korean speech recognition and synthesis system, the other 
is to show that our formalism is more describable than other 
systems employing the traditional phonological approaches\[I\] 
\[21. 
A language has its own sound patterns. By sound pattens 
we mean (1) set of sounds that occur is a language, 
(2) the permissible arrangement of the sounds in words and 
(3) the processing for adding, deleting, or changing sound 
\[3\]. 
A speech recognition/synthesis system also needs a 
phonological knowledge base system (data base), and the rules 
which govern the sound sequences and the phonological 
processes. 
Traditinoally, a computational phonological systems have 
been described and classified in terms of 'Generative 
Phonology\[4\]', However most of those approaches are not 
sufficient to describe the various knowledge of a 
computational speech system. 
In this paper, we show the feasibility of phonological 
knowledge base system with a uriification-based formalism, 
and illustrate the flexibility of implementation and 
representation. 
(closed) syllable 
(open)j syllable~ 
onset necleus coda 
\[consonant\] \[vowel\] \[cononant\] 
1 1 I 
-7 1- "7 
Fig. 1 Korean syllable structure 
3. Korean Phonology Structure Grammar 
As mentioned above, KLSG\[5\] is a new grammar theory 
for the Korean language and follows a unification-based 
grammar such as GPSG\[6\] and HPSG\[7\]. An outstanding 
characteristic of the KLSG, which has its descriptive 
capability, can be described in syntatic and morphological 
knowledges as well as "knowledge of writting system, 
phonological and semantic knowledge with a unified grammar 
theory. In this paper we are only concerned with the 
phonological structure of the Korean language. 
3-1. The phonological feature system in KPSG 
All the Korean phonological categories of KPSG are 
presented by the sets of feature and they consist of feature 
and their values. In the following, we briefly described the 
phonological feature structure of KPSG. 
2. Syllable Structure of Korean Language 
A spoken language is not only a linear of sound 
segments, but is a group of vowels and consonants to form 
large units of a sound such as syllables, words and utterance. 
In this section we present the nature and properties of 
the Korean phonetic structure : Korean syllable. The Korean 
syllable structure has two types: one is the type of consonant 
and vowel group(CV type : 71- : ga), and the other is the 
type of consonant, vowel and consonant group(CVC type : 
7-1- : gak). ~\[he Korean language, however, has a predominant 
writing system : HANGUL which is based on sound segments 
and syllables. F_zpecially, a HANGUL syllable corresponds 
to one character which is a combination of two or three 
sound symbols. In other words, a spoken syllable corresponds 
to a written syllable as a one-to-one fashion. 
Structurally, a syllable may be divided in to three parts : 
oneset, necleus and coda. The most prominent part of a 
SYL(Syllable) <--- {Consonant-Onset, 
Vowel Consonant-Coda}; 
This feature presents the consonant and vowel according 
to the articulatory properties of phoneme. 
PDF(Phoneme Distinctive Feature) 
<--- {Nasal, Voiced .... }; 
This feature presents the practical phonetic values of the 
Korean language. 
FOLL(FOLIow) ; This feature can take a set of 
categories as its values : the value corresponds to a list of 
categories. 
3-2. The Korean syllable structure rule 
We can de~ribe the Korean syllable struture of Fig 1 
using the value of FOLL features as follow : 
76 i 
(1-,1) open syllables --> 
Consonant-Onset + Vowel 
(1-2) closed syllabics .--> 
Consnnant.-Onset + Vowel + ConsonantoCc×la 
(1-l') \[SYL v ; FOL\[, { }\] -..> 
\[SYL e-o ; FOLI, { }\] 
\[SYL v ; FOLL {c-o}\]. 
(1-2') {SYL c-c ; FOLL { }1 -> 
\[SYL v ; FOLL { }\] 
\[SYL e-c ; FOLL {v}\]. 
A careful look at the features specified in (1-1') and 
(1-2') will reveal that the ocurrences of the syllabic feature 
are predictable with the form of the rules; the syllabic feature 
appears on the set of feature similar to the left hand side 
among the right hand side in the rules. 
In the KPSG, we assume the syllable structure rule as 
the general rule as follow : 
(1-3) M --> C tI, M 
C H 
M' --> M \]-I M' 
/.j" 
M /" 
C H 
It 
where M and M' are the mother categories, C and H 
are the daughter categories, in which C is the complement 
category, and tt is the head category, respectively. Since 
(1-3) is a binary., tree, we u~ (1-1') to get a syllable structure 
for (1-2'). its content is determined by the principles discussed 
later. The head in a rule plays the role of a primitive in 
the knowledge processing of KPSG, and the feature of head 
is applied by the following principle, 
A. Head Feature Principle(HFP) 
The values of the idivklual HEAD features of the 
mother are indentical to the values of the HEAD features 
corn~'sponding to its head. 
This principle guarantees that the values of these head 
features at the M(M') m×te and those at H no×It in the 
(1-3). The general idea behind the formulations is to state 
certain constraints regarding the head features, qhe head 
features include SYL and PV. 
But tim HFP does not give necessary assignment of the 
values to the syllable's element. The rules for the sound 
sequence and the phonological processes are handled by 
another principle called subeat feature principle. 
B. Subcat Feature Principle(SFP) 
'\[he value of the FOLL feature of a head is indcntical 
to the FOLL feature of its naother except for the category 
which is identical to its daughter. 
The sulx~lt feature principle states the rdationship that 
hol&; belween the FOLL vahte of the mothcr and that of 
the head in a given local phonology tree. 
C. Binding Frature Principle(BFP) 
"\[\]fis principle states the distritmtion of binding features 
over a given syllabic tree. A binding fealure is a feature 
whose value is determined with respect to a category I~r~sibly 
seperatcd by a number of syllable boundaries. "Ilfis feature 
plays a crucial role in describing the phonological process: 
vowel harmony and so on. 
~ihese principles are the constraints to select the adequate 
local syllabic tree among the inadequate local syllabic trees 
generated by the alxovc phonological rules: M ---> C H and 
M' ---> M H. The information of the basic feature values 
is prepared as the concrete phoneme entry, and is propagated 
a local tree to another local tree during the phonological 
processing, qtm operation which determines the structure cff 
feature and the sets of feature is called unification. 
The feature system in phonological knowledge base 
system plays a role of the declarative feature knowledge and 
the unification applied by the principles correspond to the 
procedural knowdedge which specify the rules governing the 
sound sequence and phonological process. From the viewpoint 
of sound inventories, each rule is described deelaratively as 
a constraint condition which is ~ttled among the features. 
"1tin equality and partiality among the feature are defined 
by the concept of unification. 
4. Pratical examples 
In this section we illustrate how the phonological rules 
in KPSG are operated. The processes of sound change vary 
in the degree of naturalness. The~.~. are various type.s of 
assimilations such as (1) nasalizal~0n, (2) palatalization, 
(3) assimilation and (4) vowel assimilation and ~ on. Among 
those various types of assimilation, we investigate the most 
colnmon type of tim nasal assimilation in the Korean 
language. 
(1-4) nasal assimilation(nassalization) 
a. k - - > ~/ ..... \[+na~l\] 
b. p - - > m/ ..... \[+nasal\] 
c. t - - > n/ ..... \[+nasal\] 
(1-4) is the formal descriptions of the phonological rules 
based on the theory of generative phonology. The notation 
k - > / - - - \[+nasal\] is to be read as "k is re-written 
as if nasal phonetics is immediately to its fight". ~lhe~ rules 
correpond to the procedural knowledge of phonological 
process. In comparison with this, unification based formalism 
is declarative knowledge representation. For example, 
(1-5) \[SYL c-o ; FOLL { i ; PV a+/3 \] / \ 
{SYL c-e; FOLL {}; PDF nasal; PVa\] 
I 
c~ \[SYL c-o; FOLL {SYL e-c; PDF nasal}: PVI3}\] 
I 
/3 
(1-5) presents the general rule form for na~l assimulation 
and here PVc~+/3means that phonetic values are composed 
according to the comtx.'sitional phonology. Now, we show 
the scheme of the rule form (1-5) with a praefial example 
in the case of rule (1-4a) 
In the (1-6), the feature's values such as c-'or v, nasal, 
and m or n, etc. colxespond to feature, SYL, PDF and PV 
respectively. At each node, the feature's values of the mother 
are passed to the head according to the head feature 
principle, the syllable's Lxotmdaries and agreement between 
the fcatures are recoglrizcd and sound's value is synthesized 
by the feature principle. 
2 '7 7 
(1-.6) ~-~- (m.k-nun) .... > ,~-~ (m -nun) 
\[SYL c-o ; FOLL { } ; PV m+a+0+n\] 
\[SYL c-o; FOLL {}; PDF nasal; PV re+a+9\] \[SYL c-c; FOLL {v}; PDF nasal; PV n\] 
\[SYL v: FOLL {}; PV re+O\] \[SYL c-c; FOLL {v}; PDF nasal; PV@\] 
\[SYL c-o; FOLL {}; PV m\] \[SYL v; FOLL {c-o}; PV eq\] 
! I 
Finally, we present briefly a Korean sound inventories 
(symbolic databases of phonetics) .of the phoneme used in 
the allophone rule. 
(1-7) 
"n(k) : {SYL c-o ; FOLL { } ; PDF "; .... ; PV k} 
"l(k) : {SYL c-c ; FOLL {v} ; PDF ";...; PV k} 
q(Z~) : {SYL c-c ; FOLL {v}; PDF nasal ; .... ; PVZ~} 
(p) : {SYL c-o ; FOLL { } ; PDF *; .... ; PV p} 
~,(p) : {SYL c-o ; FOLL{v} ; PDF nasal ; .... ; PV p} 
(m) : {SYL c-o, FOLL{v} ; PDF nasal ; .... ; PV m} 
u(t) : {SYL c-o ; FOLL { } ; PDF "; .... ; PV t} 
u (t) : {SYL c-c ; FOLL {v} ; PDF ";....; PV t} 
u (n) : {SYL c-c ; FOLL {v} ; PDF nasal ; .... ; PV n} 
u (n) : {SYL c-c ; FOLL { } ; PDF "; .... ; PV n} 
u (n) : {SYL c-c ; FOLL {v} ; PDF "; .... ; PV n} 
u (n) : {SYL c-c ; FOLL {SYLL c-c ; 
PDF nasal} ; PDF ; .... ; PV n} 
In the (1-7), the feature set shows the phonological 
constraints of each phoneme respectMy. 
5. Feature Works 
At present it is not clear just how to classify and 
integrate all the knowledge we have accumulated about the 
natural phonological processes into the grammars of individual 
languages. We introduced a unification-based grammar 
formalism for phonology. We verify that unification-based 
grammar formalism is a very useful theory for constructing 
the computational speech processing system, similar to a 
syntactical system. 
In order to construct speech recognition and synthesis 
system, we need a number of various knowledge sources : 
acoustic, phonetics, phonology, prosody, morphology, syntax, 
semantics, and pragmatics. 'Fhe problem is how to represent 
and use those knowledges. The knowledge representation 
must keep the transparency among the knowledges with a 
unified way. Unification-based grammar formalism satisfy the 
above requirements. 
We also attemp to describe the English phonological 
knowledge base system with a unification-based formalism, 
and are implementing the Korean speech recognition /synthesis 
system\[8\] with the KPSG. 

References 

Kattunen, L., Koskenniemi, K., and Kaplan, R. M., A 
Compiler for Two-level Phonological Rules, CSLI Rel~rt, 
No. CSLI-87-108, Stanford Univ., 1987. 

Church, K. W., Phonological Parsing in Speech 
Recognition, Kluwer Academie Pub., 1987. 

Crystal, D., A Dictionary of Linguistics and Phonetics, 
Basil Blackwell, 1985. 

Chomsky, N. and Halle, M., The Sound Pattern of 
Engilsh, Harper and Row, 1968. 

Chung, H. S., Korean information Processing, Ph.D. 
dis.~rtation, University of Tokyo, 1986, (in Japanese). 

Gazdar, G., Klein, E., Pullum, G. K. and Sag, I., 
Generalized Phrase Structure Grammar, Basil Blackwell, 
1985. 

Pollard, C., Lectures on HPSG, Stanford Univ., 1985. 

Chtmg, H. S., and Kunii, T. L., "NARA : A Two-way 
Simultaneous Interpretation System between Korean and 
Japanse- A Methodological Study-", Proceedings of 
COLING '86, Bonn, 1986. 

Chtmg, H. S., Current Korean : Elementary Sentence 
Pattern and Structures, Komasholin, 1982(in Japanese). 
