The primary objective of this paper is to describe an experiment 
designed to investigate the semantic relationships between the three 
basis components of a prepositional construct: the governor, preposi- 
tion and the complement. Because of the preliminary nature of the 
experiment, only simple data processing equipment, such as the keypunch 
and the sorter, was used. The implementation of this approach on a 
larger scale, however, would necessitate the use of more sophisticated 
hardware. 
The described procedure uses Russian prepositions because, while 
working on this problem, the author was a research staff member of the 
Russian-English mechanical translation group at IBM's Thomas J. Watson 
Research Center in Yorktown Heights, New York. 
While the described procedure presents a tentative approach, 
which does not offer a solution to the semantic ambiguities within pre- 
positional constructs in Russian, it does suggest a method for exa- 
mining each basic component of a given construct in relation to other 
constructs containing different types of prepositions. 
The data used in the model was collected mainly from the Soviet 
Academy of Sciences Grau~ar and, to some extent, from the Soviet Aca- 
demy of Sciences Dictionary. Initially an attempt was also made to 
compile data from other dictionaries. It was found, however, that the 
presentation and the classification of the data was not detailed 
enough for the purposes of this study. TherefOre, only some of the 
prepositions not listed as such in the previously named sources were 
included in the experiment. The next logical step, using the arrange- 
ment of the data as shown below, should be the culling out of addition- 
al data in the case of Russian, and complete data in the case of other 
languages~ from dictionaries, concordances and random texts. Following 
various sorting patterns, the results should then be tested through 
generative processes and checked against concorded 'real life' examples. 
General Purpose 
As stated earlier, the purpose of the proposed approach is the 
establishment of patterns of sementic correlations between: 
I. Given Governor and its Preposition G(----~P (left boundaries) 
2. Given Preposition and its Complement P~--"')C (right boundaries) 
3. Given Governor and its l>reposition's G~----~C 
Complement 
These relationships can be diagrarmned as follows: 
~~ (C)sn 
bl 
where sn=sematic property of any value. 
-2- 
If either of the semantic components is found to exclusively 
govern the combination of the two remaining semantic components then 
it can be said that 
(G)sn ÷ (C)sn + (P)sn = Sx 
where 
Sx = valid semantic pattern = sum of semantic properties of 
(G)sn(C)sn and (P)sn 
Where confirmed, the established semantic patterns can be incor- 
porated into a look-up table as an additional tool for parsing pro- 
cedures and for testing the validity of class-membershlp within pre- 
positional phrases. Where not confirmed or where the patterns are not 
clear, further refinement of the semantic properties of the components 
in question must be carried out. 
Method of Procedure 
Following the classification provided by the Academy of Sciences 
Grammar (henceforth referred to only as Grammar), prepositional cons- 
tructs were coded according to the grammatical category of their 
governors and the type of relationship (henceforth referred to only as 
TR) with their complement. The following governor categories were 
recognized: noun, verb, adjective, adverb, numeral. 
Next, the governors as well as the complements were coded ac- ~ 
cording to their semantic properties from examples provided by the 
-3- 
Grammar. This semantic classification was, in a sense, arbitrary in 
that it did not attempt to map the universe of all Russian words in 
the five governor categories, but tended to accomodate only those com- 
plements and governors contained in the examples, and only if their 
characteristics seemed relevant. Thus, if the Grammar provided two 
examples of the usage of the preposition and the semantic properties 
of respective governors and complements were clearly identical, only 
one example was selected. The listing of these constructs in the five 
governor categories is given in Appendix I. 
Since, in an initial study of this type, a large number of seman- 
tic classes might tend to obscure the existence of possible patterns, 
an attempt was made to keep the number of these classes at a minimum. 
As stated earlier~ the adoption of this approach in an extensive study 
of constituents within prepositional constructs would require more 
elaborate semantic mapping. For the purposes of this study, the total 
number of semantic classes for nouns was narrowed to 24, for verbs 9 
and 6 for adjectives. (See Appendix II) 
The classification of numerals and adverbs as governors was aban- 
doned when it was discovered that, according to the examples provided 
by the Graummr, their semantic values in no instance effected the 
selection of a complement of a given semantic category. Thus the 
-4- 
# 
examples citing the usage of cardinal, ordinal and indeterminate 
numbers such as '~{ECKO~KO" showed that these governors may c0-occur 
with a complement of any class. A similiar phenomenon was observed in 
the behavior of adverbs acting as governors. It is possible that a 
more detailed study of a large corpus will reveal the existence of de- 
finite relationships. 
The total number of types of relationships (TR) abstracted from 
the Gra~mmr was 42 (see Appendix Ill). Since some of the prepositions, 
especially of the compound or adverbial type, were not provided by the 
Grammar with an example of their usage in a given TR, the latter was 
derived through transformational cross-reference from the Academy of 
Sciences Dictionary, and when that proved impossible, it was supplied 
by the native speaker of Russian employed by the Project. 
In order to fit the data for each TR on a singleiBM card (for 
easier sorting), those TRs which seemed somewhat redundant or insuf- 
ficiently documented were combined and the total number of TRs was re- 
duced to Ii. Again, while the TRs were translated literally from the 
Gramsmr (admitedly, some of the translations seem a little awkward, e. 
g. 'togetherness'), the reduction of their total number was an arbitra- 
ry arrangement aimed at simplifying the overall research procedure. 
The manner in which the 43 TRs were reduced to II is shown in Appendix 
IV. 
-5- 
D 
The II TRs used in the final analysis are as follows: 
I. Attributive 7. TR expressing togetherness 
2. Temporal 8. similarity 
3. Spatial 9. deprivation 
4. Purpose I0. conformity 
5. Causative II. comparison 
6. Objective 
In addition to the aforementioned, the following data was added 
to every construct: 
I. 
AA 
O0 
MM 
XX 
AO 
OA 
etc. 
2, 
A 
P 
X 
3, 
X 
I 
General relationship 
purely attributive 
* objective 
* modifying 
* other 
attributive with objective shading 
objective with attributive shading 
Relative position to the complement 
PNP precedes the governor 
PNP follows the governor 
no preference 
Structural restrictions 
I 
none 
preposition-complement form an idiom in a given configuration 
preposition (e.g. CO~TO,~Oc~ ~ k/~O.~6~O \[JJi~.) 
-6-- 
P governor - preposition form an idiom in a given configuration 
e.g. ( ReP3~IC ,'TO" ,o,~,~ ~,,;~y) 
4. Equivalence with other TRs (syntactic difference r semantic 
equivalence) 
I yes 
0 none 
5. Other transformational possibilities 
I yes (e.g. C~.~ ,~A CT-~.'?OM - CC.q ~/ C/--O.q~) 
O none 
6. Letter codes to facilitate file search 
(usually the first three letters of the preposition) 
When punched on IBM card the data had the following format: 
Columns 
3-23 
25 
27-31 
33-34 
36 
38 
40-41 
43 
45-46 
Data 
preposition 
case governed by the preposition in a given TR 
specific relationship with the complement 
general relationship with the complement 
relative position to the complement 
governor class 
governor sub-class 
structural restrictions 
complement class 
-7- 
(cont.) 
Columns 
48 
50 
Data 
equivalence with other constructs 
other transformational possibilities 
The final step consisted of sorting the cards on individual 
columns for detection of patterns. 
v 
-8- 
APPENDIX I 
Preposition Noun Verb Adjective Adverb Numeral 
as as as as as 
~overnor Kovernor ~overnor Kovernor 
To be read as follows: 
(gra~natical and semantic categories of governor ~ semantic category 
of complement) = Type of Relationship 
Bez (NA~A)=AT 
Blagodar4* 
Bliz (NF.B)=AT 
Cerez (NM.Q)=AT 
(VX*K)=AT 
(VX-,B) =OB 
(VX-,A)=CA 
(VX÷B)=OB 
(VX~U)=SP 
(V4,E)=SP 
(VX,R)=TE 
DI4 (NE,E)=AT (VX¢C)=0B (A2+A)=CO 
(NK+B)=AT (VX+A)=PU (A2+L)=0B 
(NC~K)=AT 
(NC ~E)=AT 
(NP-E)=AT 
(NE4B)-AT 
(NL4B)=SP 
(NL÷Q)=TE 
(NC~A)=AT 
(NF*B)--AT 
~A~E)=SP 
Do 
Iz 
~4=~ 
(VX+Y)=AT (AX*Y)=AT 
(VX*A)=OB (A3+Y):OB 
(VX-B)=SP (AI4E)=SP 
(VX*Q)=TE (AS~Q)=TE 
(VX÷W)=AT (AX+B)=AT 
(VX*V)=CA 
(VX+E)=OB 
(DX4S)=TE 
(DX~A)=SP (RX÷B)=AT 
Appendix I (cont.) 
(cont.) 
(VX*G)=0B 
(VX÷B)=OB 
(VI*H)=SP 
(VXvK)=0B 
(V3~B)=SP 
Iz-pod (NF~D)=AT (VX+A)=SP 
(NE,F)=AT 
(NL-E)=AT 
Iz-za (VX+B)=CA 
(VX~E) =CA 
(NI4K)=AT (VX~B)=0B (AX*K)=AT (DX~B)=SP 
(NI~B)=AT (VX÷A)=OB (AX,A)=0B 
(NK-E)=AT (V7+K)=PU (AX÷Y)=SP 
(NE+B)=AT (VX.6Q)=TE 
(NL+A)=OB 
(NP.A) =0B 
Krugom (NF÷B)=AT (VX*B)=SP (AX+B)=SP 
Mejdu* (NATA)=AT (VX+Q)=TE (AX+D)=T0 
(NL,B)=SP (VX*D)=TO 
(NKvA)=TO 
Mimo (VX~B)=SP 
Na acc. (NB~A)=AT (VX~Y)=AT (AX.A)=AT 
(NL+A)=AT (VX*F)=AT (AX~Y) =AT 
(NI*H)=AT (VX+V)=AT (AX-~B) =OB 
(NC+E)=OB (VX+E)=AT 
(NL~A)=0B (VX÷A)=OB 
(NL4-K) =PU (V8*A)=PU 
(NL+B)=SP (VX+B)=SP 
(NN.A)=SP (VX.H)--SP 
(NM+.)=Se ¢~X~R)=rS 
(NO÷B)=SP 
*j = - i0- 
Appendix I (cont.) 
Na prep. 
Nad 
Naperekor 
Naprotiv 
Nascet* 
Navstrecu 
Vnutri 
0 acc. 
0 prep. 
Okolo 
OT 
(NB+E)=AT (V2~E)=AT (AX4Y)=AT 
(NL4B)=AT (VX~Y)=AT (AX~E)=SP 
(NL-Q)=AT (V2~F)=OB (AX+S)=TE 
(NL~E)=OB (VX+A)=OB 
(VX+K) =SP 
(VX~F)=SP 
(VX+J) =SP 
(VX ,E ) =S P 
(VX~Q)=TE 
(NL,B)=AT (VX~Y)=OB 
(NL4Y)=OB (VX-B)=SP 
(VX.A) OB 
(VX~B)=SP 
(NE *A ) =AT 
(VX~L)=Sr 
(VX*B)=SP 
(VX÷B)=SP 
(NL~B)=OB (VX+E)=OB 
(NY*A)=AT (VX÷A)=OB 
(NF÷B)=AT (VX÷B)=SP (AX~E)=SP 
(RX+E)=AT 
(DX~B)=SP 
(NE+B)=AT (VX÷K)=CA (AX~Y)=CA (DX~B)=SP (RX+B)=AT 
(NB*B)=AT (VX~A)=OB (AX+A)=DE 
(NK+Q)=AT (VX+H)=SP (AXtB)=SP 
(NE+A)=AT (AX,Q)=TE 
(NL*E)=AT 
(NE~E)=AT 
0tnositel6no (NE+A)=AT 
~C =W 
-11- 
Appendix I (cont.) 
Pered 
Po acc. 
(NB*B)=AT 
(NL4B)=AT 
(NL4Q)=AT 
(NR+S)=AT 
(NK-B) =0B 
(NL*A) =OB 
(NB~B)=AT 
(NA+A)=AT 
(NJ@Y)=AT 
Po prep. 
Po casti 
Po napravlenih k* 
Po otnowenih k 
Po povodu 
Po pricine 
Po slucah 
Po sravnenih s 
Pod acc. (NB~E)=AT 
(NE+Y)=AT 
(VX~A)--OB 
(V2+A)=SP 
(VX~Q)=TE 
(VX.E)=SP 
(VX*Y)=AT 
(VX*K) =CA 
(VX~B)=OB 
(VX*H)=SP 
(VX~A)=SP 
(VX,Q)=TE 
(VX÷A)=OB 
(VX~L)=TE 
(VX*K)=0B 
(VX,-B) -SP 
(vx~B)=os 
(VX+A)=CA 
(vx.y) =CA 
(VX~A) =CA 
(VX+L)=AT 
(VX~A)=OB 
(VX+E)=SP 
(V2~L)=OB 
(VX+B)=SP 
(VX,R)=~ 
(AX~A) =AT 
(AX.B) =oB 
(AX4S)=TE 
(AX~E)=SP 
(AX~E)=AT 
(AX~Q)=TE 
(AX÷Y)=CA 
(AX+A)=CP 
(DX-M)=TE 
(DX*B)=SP (RX.6E)=AT 
*h =~ 
-12- 
Appendix I (cont.) 
Pod instr. 
Pod le 
Podobno 
Poperek 
Pos le 
Posredstvom 
Poverx* 
Pozadi 
Prejde 
Pri 
(NF.F)=AT 
(NL*K) =AT 
(NE +E )=AT 
(NL÷E)=SP 
(NF#B)=AT 
(NF-B)=AT 
Pro 
Protiv 
Putem 
R4dom s 
Radi 
S acc 
(NBoK)=AT 
(NF-B)=AT 
(NL+B) =AT 
*X =~( 
(VX*Y) =AT 
(VX*K) =0B 
(V6+K)=SP 
(VX+J)=SP 
(VX,E)=SP 
(VX*B)=SP 
(VX4A)=Sl 
(VX~K) =OB 
(VX-A)=TE 
(vx ~ K) =AT 
(VX+B)=SP 
(VX÷B) =SP 
(VX+B)=TE 
(VX*K) =AT 
(VX*E)=SP 
(VX+D) =TE 
(VX+A)=OB 
(VX,K) =PU 
(VX.,A)=OB 
(VX~K) =AT 
(VX-C)=0B 
(VX~T)=AT 
-13- 
(AX+E)=SP 
(AXtE)=SP 
(AX*Y)=TE 
(AX+A)=AT 
(AX*K)=TE 
(AX*B)=AT 
(DX*B)=SP 
Appendix I (cont.) 
S gen. (NL-pK) =AT 
(NL~B)=AT 
(NB,E)=AT 
S instr. (NC+K)=AT 
(NF.B) =AT 
(NF+K)=AT 
(NBvA)=AT 
(NE~E)=AT 
(NC~B)=AT 
(NX-bB) =0B 
(NL~A) =OB 
(NL~B)=AT 
(NB+B)=TO 
Skvoz6* 
Soglasno 
soobrazno 
Sootvestvenno 
Sorazmerno 
Sredi 
Szadi 
(NC,C)=AT 
(NC~E)=AT 
(NL.B)=AT 
V ace. (NF*E)=AT 
(NF~T)=AT 
(NL4E)=AT 
(NY~T)=AT 
(NF*B)=AT 
(NL+E)=OB 
(VX~K)=CA 
(VX*L) =OB 
(VX+B)=SP 
(vx+o) =~ 
(VX*B) =AT 
(VX~V)=AT 
(VX,K) =OB 
(V2*B)=OB 
(VX~E)=OB 
(VX,A)=OB 
(VX+Q)=TE 
(VX+D)=TO 
(VX*A)=OB 
(vx~A)=si 
(VX*Y)=SI 
(VX~Y)=SI 
(VX+Y)=SI 
(VX~B)=SP 
(V84B)=OB 
(VX*B)=SP 
(VXtK)=AT 
(VX+A)=OB 
(VI~K)=0B 
(VX,B)=SP 
(VX~)=TE 
(AX*E)=SP 
(AX,Y)=CA 
(,~,X,Q) =rE 
(AX÷E)=AT 
(AX+B) =AT 
(A4~B)=SP 
(AX+B)=SP 
(AX,E)=SP 
(~X-tN) =sP 
(AX*G)=AT 
(AX+L) =TE 
i 
(DX~B)=OB 
(P, XtE)=AT 
-14- 
Appendix I (cont.) 
(cont.) 
(NL~A)=PU 
(NL+B)=SP 
(NL4Q)=TE 
V prep. (NB+E)=AT 
(NL.B)=AT 
(NA÷B)=AT 
(NB,B)=OB 
(NL~B)=OB 
(NK~K) =OB 
V dele 
V oblasti 
V otnowenih k 
V otnowenii 
V prodoljenie 
V qel4x* 
V silu 
V tecenie 
Vblizi 
Vdol6 
Vmesto 
Vnutr6 
Vokrug 
Voprekl 
(NF#B)=AT 
(NF,B)=AT 
(NA.A) =OB 
(NF+B)=AT 
*q = Q 
(VX~K) =AT 
(VX+E) =AT 
(V7+K)=OB 
(V8*K)=0B 
(VXtA) =OB 
(VXeE)=SP 
(vx*u)=sP 
(VX*Q) =TE 
(VX4K) =OB 
(VX+K)=0B 
(VX4K) =OB 
(VXtQ)=TE 
(VX~K)=PU 
(VX,Y)=CA 
(VX+Q)=TE 
(VX¢B)=SP 
(VX*B)=SP 
(VX+B)=SP 
(VX+B)=SP 
(VX+A)=0B 
-15- 
(AO+M)=AT 
(A3÷N)=AT 
(AX,Y)=AT 
(AX+E)=SP 
(AX,A)=AT 
(DX~B)=SP (RX+E)=AT 
Appendix I (cont.) 
Vozle (NF,B)=AT (VX@B)=SP 
Vperedi (VX4B)=SP 
Vsled (VX*B)=SP 
Vsledstvie (VX4Y)=CA 
Vvidu (VX*Y)=CA 
Vzamen 
ga acc. 
Za instr. 
(NA+A)=OB 
(NL~K)=CA 
(NL÷A) =OB 
(NL~R)=AT 
(NL4J)=SP 
(NL+R)=TE 
(NB,B)=AT 
(NL4K)=AT 
(NL*E)=AT 
(NL+A) =OB 
(VX+A)=OB 
(VX,B)=OB 
(VX*Y)=OB 
(VX~B)=SP 
(VX*Q)=TE 
(VX.~B)=AT 
(V2*B)=OB 
(VS,B)=SP 
(VX*A)=OB 
(VX+S)=TE 
(AX*E)=SP 
(AX+A)=AT 
(AX~Q)=TE 
(AX~B)=SP (RX4E)=AT 
-16- 
APPENDIX II 
Noun_____~s 
Any noun 
Concrete 
animate 
personal 
inanimate 
objects 
mass 
location 
linear (river, road, valley, etc.) 
geographical (town, etc.) 
Abstract 
• Deverbal 
transitive 
intransitive 
capable of taking prefix NA 
Adjectival 
Nouns designating TIME 
days, year, etc. 
events (birth, death, dinner, etc.) 
Measure 
preceded by a numeral 
Inner state, emotion 
Phenomena of nature (rain, frost, etc.) 
Any inanimate, abstract or concrete noun 
A 
B 
C 
D 
E 
F 
G 
H 
I 
J 
K 
L 
M 
N 
0 
P 
Q 
R 
S 
T 
U 
V 
W 
Y 
-17- 
Appendix II (cont.) 
Verbs 
Unspecified 
Motion 
Concrete action (chop, hit, tear, grind, etc.) 
Motion in the broad sense of the term (swing, jump, 
bring, etc.) 
Motion, intransitive and concrete action 
Concrete state (sit, hang, stand~ lie) 
Concrete state andmotion 
Verb phrase 
Inner state, feeling, intellectual activity (grieve, 
read, etc.) 
X 
i 
2 
Adjectives 
Unspecified 
Concrete (capable of being perceived and unmistakeably 
ascertained visually, e.g. tall, wide, white, 
etc.) 
Abstract (denoting state, density, mood, age, 
temperatures sequence, sensed olfactorily or 
aurally, etc.) 
Qualitative (good, bad, difficult, strong, etc.) 
-18- 
2 
3 
Appendix II (cont.) 
Adjectives (cont.) 
Denoting distance or location 
Inner quality possessed only by humans or 
animals 
-19- 
APPENDIX III 
Types of TRs abstracted in the initial stage of the study: 
I. Spatial 19. Separation attributive 
2. Temporal 20. Result-cause 
3. Causative 21. Temporal attributive 
4. Purpose 22. Substitution 
5. Manner of action 23. Restrictive attributive 
6. Objective 24. Property attributive 
7. General attributive 25. Objective-restrictive 
8. Similarity 26. Restrictive in time 
9. Property 27. Similarity attributive 
I0. Togetherness 28. Designation attributive 
Ii. Possession 29. Characteristic attributive 
12. Condition 30. Container and contained attr. 
13. Adverbial attributive 31. Deprivation 
14. Quantitative attributive 32. Removal (ablation) 
15. Designation 33. Conformity 
16. Attributive of absence 34. Separation 
17. Origin 35. Non-conformity 
18. Spatial attributive 36. Comparative 
-20- 
Appendix III (cont.) 
37~ Class belonging 
38. Detached object 
39. Substitutability 
40. Advocacy or defence 
41. Transgressive 
42. Distributive 
The TRs are listed in order of their description in the Grammar. 
-21- 
APPENDIX IV 
The 42 TRs were distributed into II major TRs as follows: 
2. e~oral 
Restrictive in time 
3. Causative 
4. 
5. Objective 
Result-cause 
Advocacy or defence 
Transgressive 
Distributive 
Detached object 
Substitutability 
Substitution 
Objective-restrictive 
6. Similarit E 
7. Togetherness 
8. Deprivation 
Removal 
9. Conformity 
Non-conformity 
i0. Comparative 
II. Attributive 
Manner of action 
General attributive 
Possession 
Condition 
Adverbial attributive 
Quantitative attributive 
Designation 
Attributive of absence 
Origin 
Spatial attributive 
Separation attributive 
Temporal attributive 
Restrictive attributive 
Property attributive 
Similarity attributive 
Designation attributive" 
Characteristic attributive 
Container and contained 
attributive 
Separation 
Restrictive 
Class belonging 
-22- 
