American Journal of Computational Linguistics Microfiche 9 : 3 8
LETTERS WITH VARIABLE VALUES AND THE MECHANICAL
I N F L E C T I O N OF RUMANIAN WORDS
Minerva Bocsa
U n i v e r s i t y o f T i m i ~ o a r a
R o m a n i a
The generation by computer of written Rumanian words faces
two difficult problems: to produce automatically the numerous
alternations which modify the stem and to add the inflectional
endings, building a rich set of classes and sub~lasses. The
mechanical morphological analysls i s also complicated because of
the stem’s phonetic alternatioris.
For example, the Rumanian words
UNIVERSITATE—- / U N I V E R S I T ~ ~ I- . (university)
SE3lQC?– / SERIQSI’- . / SERIQASA–e
(serious)
PUTEA– / P_OT/PQTI ;/ POATE– — – — (may
V ~ D E A– / V~/VE_EI– — / V ~ Z U I– / VADA (to see)– —-
p r e d n t the alternations
Phonetic rules describing the occurrence of these stem
modifications have several exceptions and must inelude the
presence or absehce of stress, which i s n o t marked in ordinaryRumanian I n f Lection
experiments in mechanical translation from ~ngli-shinto Rumanian
[16] and so on. Phonetic a l t e r n a t i o n i n Rumanian has been
investigated by Lombard [Ill , Felix [7] , Juglland and Edwards
[lo] , Augerot [l’] , and others.
The preparatory work f o r our automatic l i n g u i s t i c task has
several stages :
Examine the inflection of each word.
Establish the set of phonetic alternations.
Attach a specific variable letter t o each alternation.
I n our conception [4] these are different from those of
[9, 14, 151.
Design a binary code for the variable letters, tailing
i n t o ‘ a c c ~ u n ~ t h epossibilities of the IRIS 50.
Detach morphological parameters.
Code each word.
Punch a deck of cards.
The card f i l e is the Morphological Dictionary. It is exploited
by t h e programs in various ways. Here t h e working principles of
a program to produce the paradigm ( s e t of i n f l e c t e d forms) of
each word in the Morphological Dictionaryare presented.
Ir. t h i s process t h e computer w r i t e s t h e inflected forms in
the P positions of t h e paradigm ? The stem allomorphs consti-
tute a s e t A w i t h n elements. The different distributions of the
allomorphs of A in P are described by a s e e C; of grouping f u n c t i o n sRumanian Infle~tion
spelling. N e v e r t h e l e s s , the words w i t h nonc’onstant stem are too
numerous to be considered irreqular. The method of storinq the
several al,lomqrphs of the stem fortautomatic i n f l e c t i o n misses
t h e n a t u r a l uriity of,the word.
We* have Constructed a mechanical Morphol o g i cal D i o t i o n a r y / ,
containing 2 0 5 8 written Rumanian words with a synthetic repre-
s e n t a t i o n of all these phonetic alternations. An algorithm based
on this representatlon generates tne inriecrlonai noncompouna
fbrms of these words. They, are Rumanian nouns, adjectives, and
verbs, the main part belonging to t h e basic word s b c k [8, 171.
About 45 percent of them p r e s e n t stem a l t e r n a t i o n s . 1
The algorithm whose logic was given in [ 3 ] is t h e background
of a set of programs written in the programming language ASSIRIS
for the French computer IRIS 50 and its Rumanian counterpart
FELIX C-256. The proqrams were r e c e n t l y run at trhe Territorial
Electronic Calculus Center of Tirnisoara, v e r i f y i n g the algorithm.
The s y n t h e t i c representation uses G. C. Moisills notion of
l e t t e r s w i t h v a r i a b l e v a l u e s [14, 151 , which V. Gut.u Romalo
developed [9]. The setting of our research is arcu us’s theory
of mathematical linguistics [12, 131, Diaconesculs study of word
segmentation and t h e degree of regularity [5, 61 , Domonkosls
‘1t seems t h a t in Rumanian only 28 percent or e*en less of
the t o t a l number of words have these phonetic a l t e r n a t i o n s , but
in o u r d i c t i o n a r y . reference i s made generally to the most fre-
quently used words, w i t h relative frequency above 0.22% [17].Rumanian Inflection
1. Receding. The computer reads t h e word on t h e punched
card and recodes it i n t o an internal kdde; each letter is one
b y t e . A fixed l e t t e r has zone E or F (leading four b i t s I110
or 1111); variable letters have o t h e r zones. The recoding
instruction in IRIS 50 is TRTR (translate and t e s t ) .
2. Realization. The program reads the word byte by byte.
If the zone is E or F, it writes the byte i n t o the allomorph
registers. If the zone i s less t h a n E , t h e program constxucts
a realizatidn for each aIlomorph and stores i t in the allomorph
register.
The principles that #govern the decoding of a v a r i a b l e
letter into realiiatiohs are given in [ 3 ] . A s an example, take
the rule f o r regular variable letters (zone 0, 1 … 7 ) . Each
regular v a r i a b l e l e t t e r has two realizations, and in t h e i n t e r n a l
code the zone of each realization is F. The numeric of one
realization is i d e n t i c a l with t h e numeric of t h e regular variable
l e t t e r , and the numeric of the other realization is greater by 1.
The method of encodihg p a r t i t i o n s f o r regular variable letters
i s explained on t h e next frame.
The next program stage is on-frame 43.~ u m a n i a nI n f l e c t i o n
CONFIGURATIONS FOR REGULAR VARIABLE LETTERS
E i g h t zones (0, 1, . . . , 7 ) en-code regular variable letters.
Each stem has two, three, ur four allomorphs. Each partition of
the paradigm has two members f o r a regular v a r i a b l e letter; the
numeric of t h e variable l e t t e r i s copied into the allomorphs of
the. tlrst member of the partition, and -incremented by 1 i n k o
those of the second member.
Number of Allomo,rphs
Zone 3 4
0 ac/bd
1 a/bcd
2 ab/cd
3 ac/bd
4 ad/bc
5 a/bcd
6. a cdJb
7 ab/cdRumanian I n f l e c t i o n
3. Receding. The program recodes the allomorphs into EBCDIC
by another TRTR. instruction.
4 . D i s t r i b u t i o n . The proqrdm d i s t r i b u t e s the allomorphs to
their locations in another region. The word’s grouping f v n c t i ~ n
c o n t r o l s t h e procegs.
5 . ~nflection. The program adds t h e i n f l e c t i o n a l e n d i n g s
to the right of the sfem allornorph in c o n f o r m i t y with the class
and subclass noted on the punched card.
6. P r i n t i n g , The R r o g r m condenses the empty l e t t e r arid
p r i n t s the inflected forms.
We illustrate concisely these ph-ases t’or two words trom our
Morphological D i c t i o n a r y , the verbs A PUTEA (may), and A VEDEA
( t o s e e ) . They have, r e s p e c t i v e l y , f o u r and f i v e d i f f e r e n t allo-
morphs of the stem.
Input. The c o n t e n t of t h e card i s
PUTEA’ P8UlrtA8TEh V.4 L00403
VEDEA V9E9DEA v5 0703010
8U, 19k, 8 T , 9E, and 9D are v a r i a b l e letters in the external code..
Some morphological parameters are
V verb; p a r t of s p e e c h
9
5 number of allomorphs
10
0.7 worc! length
04
03 stem l e n g t h
0 3
00 g r o u p l n g f u n c t i o nRumanian I n f l e c t i o n
1. A f t e r translation i n t o the i n t e r n a l cod’e’ t h e words are
represented in s t o r a g e as
EA 84 A9 86 F2 FO
E6 92 93 F2 FO
EA., F2, FO, and. E6 represent the fixed letters . P I E, A , and V.
84, A9, 86, 92, and 9 3 represent the variable l e t t e r s U J O , a/A,
T/T.. E&A, and D/Z. The symbol p w t l l be replaced by blank.
2 . The four or three stem letters, specified by 04 or 9 3 on
the punched card, give the following four o r five allomorphs.
Thk program decodes the Irregular v a r i a b l e letter 8 4 and
pfoduces the realizations u snd 0 (bytes F5, F6) in the a l l o –
morphs a (u) and %, c , d ( 0 ) , in accordance with a translation
table. ( 3 ) The allomorphs are translated i n t o EBCDIC.
4. The allornorphs are placed in new registers as specified
by the grouping functions 03-and 00.Rumanian. In£Zection
5 . The inflectional endings a l e added.
PU TEA, PU TEXtE, PO T, ‘POTI, POATE, PU TEM, PU TETI, PO T f
-JEDEA, VEDERE , V ~ D , VEZI , VEDE , VEDEM , VEDETI , VXD , . . .
6. The computer condensgs t h e empty ‘letter in A PUTEA and
p r i n t s theinflecfpd forrps.
The v a r i a b l e – l e t t e r , method has t h e advantage of keeping the
Q n i t y of t h e word i n t h e Morphological Dictionary and producing
t h e inflected forms correctly. At t h e same time it regularizes
t h e greatest p a r t of t h e i r r e g u l a r words. The o n l y i r r e g u l a r
verbs t h a t still remain are A AVEA ( t o have), A DA (to give),
A FI (-tobe) , A LUA (to take) , A STA ( t o s t a n d ) . . The other
so-called i r r e g u l a r verbs A BEA ( t o drink),,.A MINCA ( t o e a t ) ,
A RELUA ( t o r e t a k e ) , A USCA ( t o d r y ) , A VREA . ( t o want), and a l l
t h e other semiregular verbs belonging td t h e t h i r d conjugation
[ 5 , 141 are regular for our algorithm, and so are. t h e i r r e q u l a r
nouns sod-SURORI(sister), NORA-NURORI (daughter-in-,law) ,
OM-OAMENI (man), e t c .
The program contains 1 4 5 5 A S S I R E statements and generates
t h e i n f l e c t e d forms for all the 2058 words included in t h e
Morphological Dictionary i n 1 minute 39 seconds. I t represents
an experimental v e r i f i c a t i o n of our algorithm and may be
extended without e s s e n t i a l modifications t o a l l other Rumanian
words, coded i n t h e same way.