A METHOD OF T‘RANSLATENG SSMPLE
CHINESE TEXTS INTO ENGL§SH BY MACHINE

T3951: for “19 Degrae of DH. D.
MICHIGAN STATE UNIVERSITY
David Djen-Hsien Cheng

1963

THESIS

WW////II///l////7.’///I/l////I/’I/////I///I//I////I//l

I 1
3 1293 10474 6676

This is to certify that the

thesis entitled

A b’IE'I'E-IOD OF TRKI‘ISL'LTII‘IG SE-EPLE CHINESE
TEXTS IITI'O ENGLISH BY M'aCIIIIIE

presented by
David Djen-Hsien Cheng

has been accepted towards fulfillment
of the requirements for

Ph.D. Electrical

degree in
mnglneerlng

 

4. WV? W26?

Major professor

Date 04/1/ng 9g; /€/9 3
J /

0-169

LIBRARY

t

Michigan State

University

 

 

 

MSU

LIBRARIES

 

 

 

RETURNING MATERIALS:
Place in book drop to
remove this checkout from
your record. FINES will
be charged if book is
returned after the date
stamped below.

 

 

 

 

 

MAG]?

A MOD OF MSLATING SIMPLE CHINESE TEXTS
INIO mLISH BY MACHINE

by David DJen-Hsien Chang

In translating Chinese to English by machine there are three areas
to be considered; namely (1) dictionary storage and retrieval of Chinese
characters, (2) selection of the correct meaning or the word and (3)
syntax. In view of the magnitude of the task involved, this thesis is
confined to cover the first and third areas, for which an automatic
translating system is deve10ped. Moreover, the system is primarily de-
signed to translate simple Chinese texts where the writings do not in-

volve complicated syntactic structures .

Using the radical system, a method of storage and retrieval of
Chinese characters by machine is described. An evaluation of the radical-
oriented look-up system is made. In addition, algorithms are develOped
to translate two important word order phenomena in Chinese, the idiomatic
expression and name construction. The dictionary stbrage and look-up
program written for the cm l60-A computer and machine results of a

sample translation are also presented .

The syntax system involves the analysis of word order structures

in a Chinese sentence and the synthesis into correct counterpart in

David Djen-Hsien Cheng

English. Certain Chinese words are used in a sentence for unique pur-
poses and the translation of such cases can be processed by machine
easily. The main syntax translator is derived based on the concept of
syntactic unit (synit), which is made of word or string of words bounded
by certain relationship. Chinese syntactic structures can be analyzed
by studying the relative sequences of the synits. Detection of partic-
ular sequences of synits therefore provides the key to successful trans-
lation into English. Algorithms for the complete syntax system are de-

ve10ped, and simulated results presented.

A.METHOD OF TRANSLATING SIMPLE CHINESE TEXTS
INTO ENGLISH BY MACHINE

by

David Djen-Hsien Cheng

A THESIS

Submitted to the
School of Advanced Graduate Studies of
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Electrical Engineering

1963

ACKNOWLEDGEMENTS

The author wishes to express his sincere appreciation to
Dr. Gerard P. Weeg, Professor of Electrical Engineering and.M3th-
ematics, for his guidance and encouragement in the developnent of

this thesis.

To his parents, he is ever grateful for their constant en-

couragement and understanding.

Sincere thanks are extended to Dr. Lawrence W. Von Tersch,
Dr. Richard J. Reid, Dr. Howard E. mpbell and Mr. James P. Wang

for serving on the Guidance Committee.

-11-

CHAPTER

I.

II.

TABLE OF CONTENTS

MODIETION I 0 O O O O O O O O O O O O O O O O O

l.
2.

3.
ll.

DICTIONARY RETRIEVAL AND STORAGE . . . . . . . .

l.
2.

3.

Characteristics of a Language . . . . . . .
Goals of Mechanical Translation . . . . . .
Developments Made in Machine Translation of
Chinese to English . . . . . . . . . . . .
Objectives of the Thesis . . . . . . . . .

Nature of the Chinese Character . . . . .
The Radical System. . . . . . . . . . . .
Word Look-Up Based on the Radical System
Evaluation of the Radical Look-Up System
Idctionary Information Storage . . . .
Some Solutionr to the Meaning Problem;
a. Idioms . . . . . . . . . . . . . .
b. Frame Constructions . . . . . . . .
An Algorithm.of Idiom and Word Look-Up .

Introduction . . . . . . . . . . . . . . .
Recognition Grammar Analysis . . . . . . .
Discussion of the Syntax Flow Chart and

.Associated Algorithms . . . . . . . . . . .
an Processing "Past-TEnsers" . . . . . . .
b. Deletion of Words . . . . . . . . . . .
c. Formation of Syntactic Uhit ("Synit") .
d. Synit Sequences Involving 99' . . . . .
e. Reformation of Synit of Third Order . .
f. Other Synit Sequences . . . . . . . . .

g0 mgliSh Equvalents e e e o o e e e e e 0

Conclusion . . . . . . . . . . . . . . . .

-iii—

Page

[OH

LIST OF TABLES

Various Locations of Radicals . . . .
Frame Constructions . . . . . . . . .
Look-Up Results by Band . . . . . . .
An Illustrative List of Classifiers .
Code Assignment for Synits I . . . .

Code Assignment for Synits II . . . .

A List of Frequently Used Prepositions in
Prepositional Synits . . . . . . . . .

.iv..

11
16
21+
31
Mo
1+1

55

1o.
11.
12.
13.
1h.
15.
16.

17.

18.

19.

20.

21.

LIST OF FIGURES

Information Box . . . . . . . . . . . . . . . . . . . . .
Algorithmic Chart for Processing Frame Constructions . .
Dictionary Information Storage in CDC l60<A . . . . . . .
Algorithmic Chart for Inputting Dictionary Information .
Algorithmic Chart for Idiom.and Word Look-Up . . . . . .
Syntax Flow Chart . . . . . . . . . . . . . . . . . . . .
Algorithmic Chart for Detecting "Past-Tensers" . . . . .

Algorithmic Chart for Deletion of S’] Preceded by
Adjective . . . . . . . . . . O C . . . . . . O O . . .

Algorithmic Chart of Deletion of Classifiers Preceded
by Mjectivaa]. Pronoun O I O O O O I O O O O I O O O O 0

Formation of Synits of First Order* . . . . . . . . . . .
Formation of Synits of Second Order . . . . . . . . . . .
Formation of Synits of Third Order . . . . . . . . . . .
Algorithmic Chart for Forming Nbun Synit of First Order .
Algorithmic Chart for Forming Neun Synit of Second Order.
Algorithmic Chart for Forming Neun Synit of Third Order .

Algorithmic Chart for Translating a Chinese Phrase whose
Sequence of Synits is [V] [Nll I 9? ] [N12 . . . . . .

Algorithmic Chart for Precessing Synit Sequences Formed
by #9 Preceded byaNoun or Verb Synit . . . . . . . .

Algorithmic Chart for Re-formation of Nbun Synit of
Third order 0 O O O O O O O O O O O O O O O O O O O O O

Algorithmic Chart for Translating a Chinese Phrase whose
SW11) sequence is [P] [V] [N] e o a e e e e e e e e e e

Algorithmic Chart for Precessing Comparative and Superla-
tive Words in Chinese . . . . . . . . . . . . . . . . .

Algorithmic Chart for Processing Statement of Comparison.

-v-

Page

13
18
20
21

23
28

3O
33

3h
36

37
38
1+3
AA
1+5

so
51
52
56

58
59

LIST OF APPENDICES

The Ha ical List 0 I O O O O 0 O O O O O O O O O O O
.A List of 1500 Chinese Characters . . . . . . . . .

.A Short List of Chinese Words Compiled According
to the Radical. system 0 O I O O O O I O I O O O 0

Classification of Part-of-Speech of the Chinese Word

Look-Up Program.and Sample Results Using
cm l60-A computer 0 e o e a e e e e e e e e e e 0

Simulated Translation of Chinese Passages into English

-vi-

61
63

71

71+
79

INTRODUCTION

Language as a.means of communication is possibly the most influen-
tial factor in the progress of mankind. But expansion of world wide com-
munication has been growing so, that it now reaches a point where the
chief asset of languages, that of enabling free exchange of information,
now becomes an obstacle. As the volume of information in different lan-
guages multiplies many fold each day, it is inefficient to trust the trans-
lation process to human beings alone. The advent of digital computers
opens possibilities toward logical processing of translation in terms of
analysis of the source language and synthesis of the target language.

This thesis shall deal with two phases of mechanical translation of

Chinese texts to equivalents in English.

1. Characteristics of a Language

A natural language is not develOped according to a systematic
plan. Certainly it does not attribute its growth to the highly stylized
structures of logic and reasoning. Rather, a language is a convention
that has evolved through centuries of use. It must constantly absorb
new ideas and create new words and meanings. Indeed, as a tool of come
munication to reflect multitudes of facets, languages are develOped

based on many complex patterns.

-2-

2. Goals of Mechanical Translation

The ultimate goal of translating languages by machine such as from
Chinese to English is to input to a computer a passage of Chinese and
receive as output a lucid English counterpart of the input Chinese. To
relate closely with the actual problems, we should aim.to accomplish the
following:

(1) Construction of a complete recognition system for
Chinese, both in meaning and syntax.

(2) DevelOpment of effective methods to translate
Chinese sentences into corresponding English
automatically.

(3) DeveIOpment of the algorithm and principles of

programming for the translation process.

As the problems listed are very much interconnected with each
other, the fundamental task becomes that of method of attack. An ideal
starting point would be to conduct an exhaustive study of the Chinese
language structure, but this is obviously impossible. Furthermore, the
established results on Chinese by linguists, compared to the magnitude
of the problems, have been meager. In fact, there is yet no established
grammatical system that can meet the fundamental requirements of machine-
oriented analysis - that of a formal ordering of the structure of Chinese.
This lack of linguistic elucidation, mainly due to the fact that lin-

guists have been more concerned with methods of teaching rather than

analyzing the language, certainly does not relieve the existing critical

situation.
3. DevelOpments Made in Machine Translation of Chinese to English

A recent report from the National Science FoundationIl] shows only
a handful of organizations, all in the U.S.A., that are actively involved
in machine translation work from Chinese to English. The major effort
in this field comes from the university of California, International
Business Machines Corporation, and Ohio State university. At the
University of California work has been concentrated on compilation of
a set of cross-reference indices for 10,000 Chinese characters, and
publication of the results is said to be forthcoming. I.BaM. is also
engaged in solutions of encoding problems as well as develoPing storage
devices for the dictionary. The work at Ohio State university is in-
volved with some preliminary studies of the syntactic structure of
Chinese. However, other than the mentioned publication, very little

published results are in evidence so far.
A. Objectives of the Thesis

In view of the aims and what has been accomplished it is recog-
nized that machine translation work is a formidable task. Indeed, it
is so difficult that one can attack only a few isolated problems at a
time. But in developing an automatic translator for Chinese, several

basic questions must be considered. These are:

How is the Chinese text to be input?

NNst the input be performed by one skilled in

Chinese linguistics?

Can the computer not only discover the correct
meaning of each word but also arrange a.meaningful

English translation?

To accomplish a.meaningful translation the trans-
lator must not only use a dictionary to find the
meaning of a word, but it must also examine the
context in which the word appears. Can a computer

do all this?

As a partial solution to some of these problems, this thesis will
handle two situations. First, the computer will be made to perform.a
dictionary search for all the English equivalents of Chinese words,
producing not only meanings but also grammatical information. Second,
the computer will be presented with the correct English equivalents of
the words in a Chinese text and their grammatical uses, and will output

a correct translation.

It should be clearly understood that these are two distinct pro-
cesses. The first process, the dictionary look-up, requires an input of
Chinese words. These words can be input by any well-trained clerk, who
needsunderstand no Chinese. The output could conceivably be provided
to one familiar with English and only mikfly familiar with Chinese, who

could then sort out the correct meanings of the passage. In the second

process, the input information to the machine is highly refined, and
includes the correct meaning, part-of—speech and other potential syn-

tactic functions of the words. The machine then will proceed to output

the English translation.

Clearly there is a gap between the two processes and one which
will eventually be closed. It is visualized that a program using both
the dictionary look-up and the passage translator should be provided
which will on a statistical basis select a.meaning for each word in a
passage; then attempt to translate the passage on the basis of the word
meanings selected together with their parts-of-speech. Then an English
speaking person will look at the result and declare to the computer
that the result was good or bad. If bad, the computer re-selects the
word meaning and tries again. The principle to be employed is similar

to the learning machine of Arnold [2].

The translating system in this thesis is designed to translate
fairly straightforward Chinese text. The main reason is that these
unritings have more orderly syntactic structures. The two books con-
Estantly used as references are Chinese Reader For Beginners by Shah
Wing Chan [3] and Character Text For Mandarin Primer by Yuen Ren
Cflhao [A]. .These two books are primarily used by students in beginning
Cklinese. The attributes of the texts chosen are that they are gram-
martically correct; require a relatively small vocabulary; and they do

ncrt involve complicated structures. Thus the translator will not be of

-5-

such a complexity as to overpower a computer; yet on the other hand,
the translator will be handling reputable passages which occur in the

standard active use of the language.

A.word should be given concerning the programs written. The major
contribution of the thesis is not so much the programs as it is the al-
gorithms from which the programs sprang. Given the latter, the former
can be implemented by any clever programmer. To date, the dictionary
program for idiom.and word look-up is written and runs. Samples of
results of that program.are included with the thesis. The second pro-
gram, the syntactic translator is in the code checking phase, and no
results are available. However, in a thesis of this sort, results con-
stitute the "proof of the pudding". Thus, in lieu of the machine re-
sults there is included a translation by hand of several passages in
Chinese, obtained by strictly following the algorithms presented in

Chapter 2.

I. DICTIONARY REERIEVAL AND STORAGE

1. Nature of the Chinese Characters

Characters or words of the Chinese language derive their present
forms through several methods. Two comon ones are the pictographic
and the ideographic forms. The mixture of sources from which the
characters are derived probably leads to the fact that the language
does not have a basic system, such as the alphabets in English, by
which characters are formed. Rather, Chinese words are unique in con-
struction; through combinations of strokes the possibilities of forms
are literally inexhaustible. Since words must be fed into a machine in
some code best suited for recognition of the written language, the input

of Chinese text to the machine consequently presents a critical problem.

2. The Radical System

During the past thirty years Chinese linguists have come up with
almost one hundred different look-up systems . Unfortunately, none has
really proved to be superior enough to win any great amount of favor from

its users .

Investigations on a system called the radical system has shown that

it has features which may eventually prove it to be an effective and unique

-7...

look-up system for the Chinese characters. A radical is a character

form on the graphamic level somewhat the parallel of alphabetic symbols

in English. A.Chinese word or character which is not itself a radical
consists of two parts, the radical and the auxiliary. The radical is

one of the list of 21k radicals (Appendix 1). The auxiliary is the re-
maining portion of the word, which sometimes can be another radical. In
order to look up a word using the radical system, one would first identify
the radical associated with the word, and then by going to the list under
that particular radical the word in question can be located. The latter
step should present no problem unless the list is quite long. In that
case, one can divide the list further by the number of strokes each word
takes. Resorting to this scheme is not a complicated procedure since

the average number of strokes of a Chinese word is approximately ten;
which means the counting process, when used, would not be a time consuming

job.

Once the radical-oriented vocabulary is compiled, the digital
equivalents of the words used as input to the machine can be formulated.
The numbering of the digital equivalents shall be decided with special
attention paid to the size of storage available in the computer; such a
process therefore has to be done arbitrarily. It is conceivable to cor-
respond the equivalent number of the word with the memory location at
which the dictionary information for that word is found. Appendix 2 con-
tains a Chinese vocabulary of 1,400 commonly used words, and the listing

is based on the radical system, A smaller vocabulary, to be used for

demonstration purposes later, is shown in Appendix 3, and an arbitrary

digital equivalent is indicated alongside each word.

3. Word Look-Up Based on the Radical System

Word look-up is thus a series of matching processes. Prior to any
machine input one would first locate the word to be translated from the
radical list. Then the digital code of that word is used as input to the
computer, which in turn proceeds to locate the section of the memory con-
taining all the information on grammar and meanings of that word. This
information is then placed where needed for the next stage of the trans-

lation process.

As an illustration, assume that a portion of a certain Chinese

passage to be translated contains the words

it E7 32 3

From Appendix 3, these A words can be located under the radical lists of
‘x , [3 , r'-7 and J respectively. The digital codes for them are 211-70,
1730, 2220 and 1100, which are to be input to the machine as unique
equivalents to the h words. The machine recognizes these h digital codes
and proceeds to bring out from.memory information on the words. It can
be seen that if the digital equivalents are exactly the memory locations
of the words the retrieving step is made easier. A detailed description

of programming procedures using the CDC 160-A Computer is shown in Section 7

-10-
of this chapter.

A. Evaluation of the Radical Look-Up System

The look-up system so described does not require the input operator
to be familiar with the Chinese language. A person with some intuition
on symbols can be trained to recognize the set of radicals. This step of
human look-up prior to machine input seems to be unavoidable, at least

until automatic pattern recognition schemes are further developed.

While it is ideal that Chinese characters can be looked up by any
"symbolaminded" person, efficiency and uniqueness are other essential
qualities to be considered in a look-up systems The radical approach
gives a unique system, but not a very efficient one. First, 21h radicals
are excessive. Next, inconvenience arises from the fact that there is no
standard for pinpointing exactly the placement of the radical in the char-
acter. The list shown on the next page, illustrates the various positions

in which a radical can be found in a word.

A recent publication by Tu Hsui-Chih [5] indicates his efforts to
revise the radical forms so that the final list contains only 85 radicals.
This is a reduction of 129 from the original one and definitely takes a
giant step toward a simplified systems Along with the reduced list, Tu
is working on a look-up system of characters by their structural composi-
tions. A word is to be categorized into one of five classes: upper-lower,

left-right, outer-inner, middle-side, and major-minor. This partitioning

-11-

Table I Various Locations of Radicals

Radical in upper part of word: 4. in #57
lower part of word " in JL

left part of word ;; in 123

right part of word 1. in til

middle part of word I3 in 3;

Radical split in the middle part of word: in in g
Split in the middle part of word 4‘3“ in {5;}

split in the upper part of word 99 in éfii

split in the lower part of word /\ 1J1 it

Radical in upper-center part of word: ii in 3;.
lower-center part of word .§— in fik

left -center part of word :9 in .%§
right-center part of word ii in ii,

upper-left part of word J; in 13.

lower-left part of word 55- in it...
upper-right part of word {7 in §8
lower-right part of word a, in as.

of characters provides a means of look-up free from any confusion due to
the difficulty of locating the radical in a character. However, all

these works are still far from any fruitful result.

5. Dictionary Information Storage

The role of a dictionary is to store all the information pertinent
to the meaning of the word and its potential functions in a sentence.

This variety of information is to be divided into several fields. The

first field shall contain all possible meanings of the word, either by
itself or combined with other specific words. The codes of the meanings
are actually represented by the locations of their equivalents in English.
The parts-of-speech of the different meanings should also be stored in
their respective order. These parts-of-speech constitute the most essen-
tial grammatical information of the words, and consequently are extremely
useful at the stages of'meaning determination and syntax. The classifica-
tion of the part-of-speech must therefore be as refined as possible so

as to allow precise analysis of syntactic structures. Appendix E lists
all the parts-of-speech and their sub-fields. This classification is

not far different from that usually found in a.dictionary. One notice-
able change is that of shifting a number of words into a category called
special terms, which are commonly used words or groups of words whose
grammatical roles are too complex to be classified into single part-of-

speech categories.

Lastly, one other field must be incorporated to include codes which
reflect the potential function of a word in a sentence. This function
supplements the part-of-speech to serve in the capacity of detecting word
formation in the context of the word. .More will be said of this in the

chapter on syntax.

The method of storage of the dictionary infbrmation can only be
decided by the particular memory available. The conversion into bit
forms is simple, but the word length of the computer governs the amount

of information that can be stored in each word. Generally, one may

-13-

conceive of an information box to associate with each meaning of a word.
The number of information boxes then equals the number of'meanings a word

can have. The box content is divided into 3 fields as follows:

 

 

 

 

 

 

Meaning Part-of- Potential
Syntactic
(i.e., Location of English Equivalent) Speech Function

Figure 1 Information Box

The number of meanings of words is obviously varied. Then the storage
space allocated for each word should also be varied accordingly, and a
means of indicating end-of-list of information boxes of the word is

necessary.

From here on strings of Chinese words will be more often considered.
A string always ends with a punctuation mark. For uniformity the rep-
resentation for a string of words is designated by:

first word of string

kth

a.punctuation mark.

W W ... W ... Wfi, where W

l 2 k l

word of string

5'!
’d W:
II

The three fields with respect to Wk are abbreviated by

LOC(Wk) = location of English equivalent of Wk
POS(Wk) = part-of-speech of wk
FUN(Wk) = syntactic function of wk

while the computer word is referred to as

INFO(Wk) = total content of information box of Wk.

This nomenclature will become useful in later sections.

.1h.

6. Some Solution to the Meaning Problem

Determining the correct meaning of a word from its many possible
meanings calls for a thorough contextual analysis of the word. Usually
a skilled linguist is required to make the appropriate choice. However,
there are two types of word construction in Chinese which are strictly
dependent on the rigid order of specific words, and they are (l) idioms
or combination words and (2) frame constructions. The meaning of such
entities is unique, i.e. if a particular phrase can be recognized as an
idiom or a frame construction, its meaning is fixed. Thus, one can
incorporate these constructions into the machine program to allow auto-
matic processing and thereby reduce the problem of multiple meaning to

some extent.

6a. Idioms

In Bar-Hillel's [6] words, "A given sentence in language Ll is
idiomatic with reSpect to a language L2 to a given bilingual word dic-
tionary from Ll to L2 and to a given list of grammatical rules if, and
only if, none of the sequences of the L2 correspondents of the sequences
of words of the given Ll sentences is found to be grammatically and
semantically a satisfactory translation after perusal of the applicable
grammatical rules." Idioms can be removed by extending the dictionary.
Their specific patterns of word orders can be stored as part of the

word information, and these patterns can provide the necessary diagnostic

-15-

clues to insure detection by machine.

For example, the English translation of the word.1§? is one of
(1) important, (2) to want, and (3) must. We find that in every case
when 2? is immediately followed by the word 7% , which means (1) to

be or (2) right, then the correct meaning of 3? fig together is "if".

The digital equivalents of {g and 79,; in Appendix 3 are #130 and
2500 reSpectively. The code 2500 must be part of the dictionary informa-
tion content of the word 3? (4130). There may be other word or words
which when preceded by g? form a single meaning, and the digital codes
for those words should also be pre-stored along with that of -§E .
Therefore, as soon as the word g? is looked up, the digital code of
the word following i? is also brought in to match with the pre-stored
numbers. In the case of' %? f5 y,‘the match.is satisfied and the cor-

responding information box for the meaning of $ 19: in combination re-

places that of the word é? , and the word ii, need not be looked up.

6b. Frame Constructions

Frame constructions are so named because its members are located
at the beginning and the end of a clause separated in the middle by a
word or string of words, which is thus framed by the elements of the
construction. These constructions usually indicate different types of

spatial or temporal relationships. The initial element of a frame

-16-

construction is a verb-preposition. The closing element is a preposition
or a noun with spatial or temporal connotation. Following is a list of

the most commonly used frame constructions:

Table 2 Frame Constructions

SPATIAL FRAME CONSTRUCTIONS:

at (to be at) up = amid
J; = above . . .
'1: = below . . .
“at, = in front of . . .
33‘ = behind . . .
13.7 = among . . .
Db = inside of . . .
9+ = outside of . . .
”a = in, inside . . .

TMORAL FRAME CONSTRUCTIONS:

‘35? (to be at) . . . 3% {CE (time) as when , , ,
ii] (to, at) . . . 9,} 1.; (time) = at the time of . . .
g; . . . 3% 1.; (time) = when . . .

Indicators of the frame construction are assigned to both the
initial and final elements. For instance, take 4‘; ... 8-343; (when
...) in a clause; the machine first detects the initial element indi-
cator in the word at , and proceeds to find the final element indi-

cator within the clause. In this case it is found in 3%“: , and then

-17-

the list of all possible frame constructions, which has been pre-stored
in the machine, is consulted. Eventually the particular combination of
words (i.e., i; and 3% {i ) is found in the list and the meaning of

“when"-and its associated information take the place of the meaning of

in ,and 8&1»; is deleted.

Sometimes similar word formations prove to be a false frame con-
struction. The clue lies in members interior to the frame construction.
For example, i (at) 9%“: (time) in the list shown would mean
"when ....". Consider now it E5: 541-1»; , where é fl means "what", this
phrase then should be translated "at what time", or in the natural order
in which the words appear. Words (like a )fi- constitute the clue, and

will be called a frame stopper.

During machine processing, the indicating signs for frame con-
struction can not be found by examining the parts-of-Speech of the words.
But these signs can be incorporated into the category of potential func-

tions of the words. Assuming the following arbitrary codes:

11 for frame construction initial element
12 for frame construction stOpper

13 for frame construction final element,

the flow chart describing the automatic processing is shown in Figure 2.

-18-

Using the nomenclature of Section 5 of Chapter 1, we assume the

string of words to be examined is represented by W1W2 ... Wk ... Wb.

 

 

 

 

 

 

 

 

Load INFO(Wk)

 

 

 

 

 

 

Pos (wk) 4. 9x Yes

 

 

 

Next Stage

 

 

No

[Ki-l -> K NO

 

 

 

 

 

 

 

FUN(Wk) .3. 11

Yes

 

 

 

 

 

K + l ——> K

 

 

 

 

 

 

 

 

 

 

 

 

Pos(w)3.9x Yes ”

 

 

 

 

Yes

 

 

 

 

 

 

 

 

 

 

FUN (W);l3 NO

 

 

Consult List of

Frame Constructions

Figure 2 Algorithmic Chart for Processing Frame Constructions

-19-

7. An Algorithm of Idiom.and Word Look-Up

The system of word storage and look-up designed for the CDC 160-A
computer is now described in two parts. The first part deals with stor-
age of dictionary information into memory. The second part describes

the look-up procedures.

Each word needs ten octal positions to store its information, which
consists of two idioms involving the word and one meaningf. For illustra-
tion take the word .135. ; from the radical list under-ii- in Appendix 3, if;
is identified to be numerically equivalent to 2150. The twoaword idioms
involving i??? as the first word are {3} ii and $3 to . The numerical
equivalents of {a and.ig are 3050 and 3030 respectively, and the loca-
tions of the English equivalents of the idioms are 5110 and 5020 reapec-
tively. Both idioms are nouns (numerical code 12), and have no signifi-
cant syntactic functions. The meaning of $33. by itself is "to learn"
which is a verb (numerical code 41) and whose English equivalent canbe
found at location 3020. Hence, the memory locations 2150-2157 inclusive

shall store the information associated with~€§» in the following fashion:

 

*For convenience, we shall do away with the multiple meaning problem by
incorporating only the correct meaning of the word in its context. This
does not hamper the basic principles of storage and look-up since any
additional meaning of the word merely increases the memory space but the
storage and look-up formats are similar. moreover, the restriction to
two idioms clearly does not alter the basic process; but for a small
computer, such as the CDC 160-A, does make the memory go farther.

-20-

 

 

 

 

 

 

 

 

 

 

 

 

 

2150 3 o 5 0 Digital Equivalent of It}:

2151 0 0* 1 2 Syntactic Function and Part-of-Speech of $3.81
2152 5 1’ l 0 IocationofEnglish Equivalent of Idiom E?) fl
2153 3 o 3 0 Digital Equivalent of is

2151+ o o l 2 Syntactic Function 8. Part-of-Speech of £35.43
2155 5 0 2 0 Location of English Equivalent of Idiom fig; #3
2156 0 0 h l Syntactic Function and Part-of-Speech of a}
2157 3 0 2 0 Location of English Equivalent of Word 2%.;

 

Figure 3 Dictionary Information Storage in CD0 l60-A

The memory section of 1000-W77 inclusive in CDC 160-A is allocated
for storing dictionary information of all the words in Appendix 3. This
information must be stored sequentially, that is to say, storage begins
at 1000 with the dictionary information of the word - , continues at
1010 with the word 3‘» , and so on to 11-770 with the punctuation mark ! .

An end of list sign is necessary, and an 0077 is chosen for that purpose.

The algorithmic chart for inputting dictionary is shown on the following

pasc-

 

«x-
_ 00 designates no syntactic function.

-21-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

K = 1000 I
l
—-)| Input 1+ octal digits
Is Input YES Next
End of List (0077): . stage
No

 

 

Store input into K

 

 

 

 

 

K + l -———e> K

 

 

 

 

 

 

Figure A Algorithmic Chart for Inputting Dictionary Information

-22 ..

The look-up process begins with finding the equivalent digital
codes of the words to be translated. The method is outlined in Section 3
of this chapter. Suppose the Chinese passage below is to be translated,
then from Appendix 3 the equivalent codes are found and indicated directly
below the correSponding words.
if a”. in 11 ii is iii #3 i i a 9%
2&70 10h0 3550 h3h0 A320 1230 2150 3030 2000 3h00 4670 2750

ii a m -
12u0 3660 3110 A710

 

These numerical codes become the input to the machine. As end of list

sign is again needed, and an 0077 is again used.

The first step of machine look-up is the search for idioms. If a
combination of adjacent words forms a single meaning, it must be found
at this stage; for otherwise the single meanings of the words will be
output. It should be noticed that the problem of multiple meaning
actually lessens with the increasing number of idioms in a.passage since
these idiomatic meanings are unique. The procedures of idiom.look-up is
described in the example in Section 6a. The algorithmic chart for the

combined idiom and word look-up is shown in Figure 5.

The routine to search frame constructions can be conveniently
inserted after the idiom and word look-up. The combined program.of dic-

tionary storage and retrieval written for CDC l60-A computer is shown in

 

 

 

Input h Octal Digits

 

J,

 

 

 

Store in Temp. Cell "m"

 

 

 

 

-23-

 

 

 

 

 

 

 

 

 

Yes

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Yes
Is Input End of List (0077)?
. No , -
Input A Octal Digits
Store in Temp. Cell "n"
l; Yes Add 3 to Content of m
Is Input End of List (0077)? ‘
No _ .
Add 3ito Content of m
content Of ? Content 1
Location = of n
Specified by m. Set LOOP Count = 2
l.No
Add 3 to Content of m. Output Content of Locationlé;q
Specified by m
Content of A ? Content No I
Location a ofln
Specified by m. Add 1 to Content of m
Yes
Set Loon Count = 2 l. Reduce2L00p Count by l O
2. Count e 0 lL-'
Add ,1, to Content of m
Yes
OUtPut 001113911t 0f Location Store Content of n into m.|
Specified by’m
I Content of n No
End of List?
1. Reduce LoOp Count by 1
No
42 O "—" iYes
20 .1:
Count ' Next Stage

 

 

 

 

Yes

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5 Algorithmic Chart for Idiom and Word Look-Up

 

 

-2h-

Appendix 5, and machine results of translating the given Chinese passage

are also included. The results worked out by hand are shown below.

Table 3 Look-Up Results by Band

 

 

Words in Syntactic Part-of- Location of
Passage function Speech English Equivalent
it 00 12 2&20
:5. 00 31 3h10
1k. ii 00 ’41 2700
it 00 15 6110
1:3 00 63 61+00
a? #3 00 12 5020
a 11 M 0150
E )2 12 82 7030
9.2} 7.5; 13 11 6010
5% it 00 #2 6120
. 00 91 6600

 

 

 

 

 

 

It can be seen that the machine results and hand results are identical.

II. SYNTAX

1. Introduction

Syntax is the process by which words combine to form phrases and
sentences according to certain grammatical rules. It is, therefore, that
we begin by assuming every sentence has a structure, that is, that each
string of words which purports to be a sentence is as a.matter of fact a
sentence. With the string of Chinese words and its correct meanings and
grammatical preperties supplied to the computer, the machine must be made
to process these words into prOper order for synthesis into English. This
means the grammatical contents of the words are to be manipulated while
the conceptual contents of the message remains constant throughout the
automatic processing. In general, grammatical analysis of a sentence
may aim at various results. The method to be used in the analysis may
differ considerably, depending on the particular aim at hand. One can
set up a generative grammar to produce strings of words which are sen-
tences of a given language. Another type of grammar may be aimed at dis-
tinguishing sentences from non-sentences. However, these grammars do
not analyze existing features of a sentence. For this, a recognition
grammar can be set up which takes a given sentence of a language and finds

the syntactic structures of the sentence. The analysis to be taken here

-25-

126-

is that of recognition grammar, and assumes that in one way or another
there is a decision involved concerning the syntactic relationship of

each word string.

2. Recognition Grammar Analysis

The two common methods of describing a sentence are phrase struc-
ture analysis and dependency analysis. The former, also known as the
immediate-constituent analysis, starts at the syntactic level and con-
siders a step-by-step breakdown into components of increasingly lower
order of complexity. The dependency analysis, on the other hand, breaks
the sentence up into a.tree formation with the predicate as the stemt P.
Garvin [7] conceived a method similar to that of the phrase structure
analysis with the significant difference being that the minimum.units,
the words, are taken first and these units are gradually fused into units
of higher order of complexity. This thesis follows that approach; in
this section a recognition grammar is deve10ped. This leads to an al-

gorithm for translating Chinese to English.

Similar to the process of translation by human beings, syntactic
analysis should follow that of look-up and meaning determination of words.
As Oettinger [8] has stated, word-by-word translation is a linear approx-
imation to the more SOphisticated transformations capable of mapping ele-
ments of the source language into elements of the target language. It is

generally recognized that these more SOphisticated transformations will

-27-

be functions, not of isolated words, but of words and their context.
However, Jumping from.word-by-word analysis to an analysis of contextual
influence in the broadest sense is not likely to be ideal. Initially
restricting analysis to linguistic context, and to a small neighborhood
of a word at that, is more likely to lead to results both linguistically

significant and concrete.

Therefore, the initial system.analysis is to provide for signs
that lead first to the readily determinable relations and gradually shift
to more complex relations. The initial indicating signs are derived from
the part-of-speech codes. From there, groupings of words into possible
structural, sequential, meaningful combinations shall take place in terms
of the grammatical information contents of the units already determined.
The process of recognition is not carried out in an arbitrary manner but
in a definite sequence in which completion of the work in one stage of
recognition insures the successful translation at the next stage. This

combining process is built up until the sentence level is reached.

The flow chart shown on the next page describes the procedures

involved in the syntax system.

-28-

l 2 k

t

Processing "Past Tensers"

INPUT: A String of Words W W ... W ... W? + their information

boxes.

 

 

 

 

 

 

 

Deletion of Words l

J.

Formation of Syntactic unit (Synit) of First Order I

 

 

 

 

 

 

 

Formation of Synit of Second Order I

T

WL

 

Formation of Synit of Third Order

5‘:

Synit Sequences Involving 9’]

J2

Re—formation of Synit of Third Order

J.

Other Synit Sequences

J7

Output of English Equivalents

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 6 Syntax Flow Chart

-29-

3. Discussion of the Syntax Flow Chart and Associated Algorithms

3a. Processing "Past-Tensers"

The fact that Chinese is not an inflected language leads to the
necessity of contextual analysis of verbs as to their tenses or aspects:
In one situation, single characters following certain verbs carry the
grammatical connotation of past action. There are two words, 7 and 52.. ,

commonly used in this sense, which shall be called "past-tensers."

The function of the word '3 in a sentence is many fold, but in
the case when it is used to convert a verb into or past tense or com-
pleted aspect, it follows immediately after the particular verb. Thus

oz "to eat" becomes «'5, 3’ "eaten", and E7 * "to return" becomes

{.57 i ‘5 "returned".

When ii is used as a sign for past tense it is also placed
immediately after the verb the tense of which it affects. Thus, "to
have seen" in Chinese is 75 1 £7. ("to see" %’ 2, ), and "to have
eaten" at {a .

The procedure for making verbs into past tenses is shown on

next page.

* Chinese linguists have been debating on whether verb tenses really
exist in the Chinese language. Some asserts that the word "aspect" is
perhaps more fitting to indicate the temporal order. However, the
analysis here is not involved to such depth, and the words "tense" and
"aspect" will be used interchangeably.

-30-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

K = 1
V
K + l-—¢ K , Load INFO (wk)
1
Yes
No
No

 

 

2
P08 (wk) s Ax

 

 

 

Yes

 

 

K + l -———+> I(

 

 

 

 

 

 

POS (wk) 3 85 (3',g§) no

 

 

Yes

 

 

Change Tense of Wk-l’

a Verb, to Past Tense

 

 

 

 

 

 

 

 

Delete INFO (wk)

 

 

Figure 7 Algorithmic Chart for Detecting "Past-Tensers"

-31-

3b. Delection of Words

There are words in Chinese which when translated into English are
considered excess baggage. One such case is the usage of fié} after ad-
jectives. The function of 3:) here is to carry the modifying tone which
is missing when the adjective is used alone in Chinese. Generally it is
also grammatically incorrect if the Chinese adjective is not followed by

(.343 . However this is not true in English. Therefore, {3’9 can be de-

leted in this case without any misinterpretation.

Another case of excess baggage involves certain usages of units
or classifiers. It is a rather interesting phenomenon in Chinese that
when a number is used with a substantive, a unit or a classifier must
also be included. There are various types of classifiers to associate

with the substantives. An illustrative list of these is shown below:

Table A An Illustrative List of Classifiers

The classifier for E? (horse) is 7:
" ;$. (car) is .$@
" 3 (letter) is 3%
" iii (tree) is if;
" (dog) is 41

H

(house) is fifi

Thus, "two horses" in Chinese is .2. g; 57 , and "three cars"
become __=_ *5? SF , which literally means "two classifier (horse) horses"

and "three classifier (car) cars" reSpectively.

-32-

An extension of this phenomenon is the usage of more general terms
like the adjectival pronoun "this" and "that" to describe a substantive

instead of a specific number. Here, too, the classifier must be used.

OIIP

"This (IL_ ) horse" in Chinese is EEJE.W~ , and "that (-fiP ) car" is

#7! $fiy$. In the case of Specific numbers, the classifier should be re-
tained for the purpose of clearer interpretation. But when used with
adjectival pronouns, it can be deleted because the modified terms, like

horse and car, do not require the classifiers in English usage.

Algorithmic charts for the deletion processes are Shown in Figures

8 and 9.

3c. Formation of Syntactic units

Analyzing the relations among the components of any particular
passage presupposes a prior cataloguing of both the possible significant
relations among passage components and the prOperties of representations
that reflect these relations. The relations in question are formal ones,
as between adverb and verb, or between noun and adjectival.modifier.

These relations are typically reflected in the order of the related repre-
sentations or in structural similarities of the related representations.
For example, in Chinese the adjectival modifier always precedes its noun,
and adverbs precede verbs. These formally fixed relationships constitute
the very information which is needed to construct correspondence with

the target language.

-33-

 

 

 

 

 

 

 

K + l—9-K Load INFO (wk)

 

 

 

 

 

 

 

 

POS (wk) 2 9x ———/Yes \) Next Stage

 

 

 

 

 

iNO

Pos (wk) 3:21:

JIYe s

K + l ——>- K
Pos ("1.) 3, BMW N°

JR 8

Delete INFO (Wk)

 

 

No

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 8 Algorithmic Chart for Deletion of 3‘.) Preceded by Adjective

 

K + l-SDK

 

-3h-

 

 

 

 

 

 

 

 

 

 

No

Load INFO (Wk)

 

 

 

 

7
P08 (wk) a 9x

 

I YeB/

 

 

 

 

 

(POS (wk) ; 15

 

 

 

Yes

 

 

K.+ l-———+-K

 

 

L

 

(Pos ("1.) 3 63
(classifier)

 

No

 

 

Next Stage

 

 

 

 

 

Yes

 

 

 

Delete INFO (wk)

 

 

Figure 9 Algorithmic Chart of Deletion of Classifiers Preceded by

by Adjectival Pronoun

 

-35-

Therefore, to establish the relationship among words, we set out
to formulate constituent patterns of words in terms of their formal gram-
matical prOperties. The immediately accessible ones are their part-of-
Speech codes. We begin by listing the very simple relationships. When
adjacent words satisfy any of the established relationships, they can be
bounded to form a unit which shall be called syntactic unit, or "synit"
for Short. A synit is not necessarily restricted to contain one or two
words only; rather it can include a sequence of words as long as the
adjacency relationship is satisfied. Gradually the synits of lower
orders starting with the first order Shall be deve10ped into the com-

position of synits of higher orders.

In formulating the set of rules to form synits we rely not on
linguistic intuition or pre-existing descriptive analysis but on the
study of a.large number of passages derived mainly from the two books
previously mentioned [3] [A]. This has the immediate advantage that one
can be sure, even before testing, that one has a set of rules of fairly
wide applicability. Furthermore, in dealing directly with the writings,
one gains familiarity with the problem areas that may arise and thus
is in a better position to resolve the ambiguities. Inadequacies will
inevitably occur, but in the light of accumulated results, this empirical

approach does provide a working method toward the paramount aim.

The synits of first order are formed by directly searching the

part-of-Speech codes associated with the words. Assume the following

designations:
Noun- synit of the first order - (N)
Adjective- " - (A)
Verb- " - (V)
Preposition- " - (P)
Number- " - cg)
Conjunction- " - (C)
Special term- " ' (S)
Punctuation mark- " - ($)

The combination of words, each of which is represented by its
part-of—Speech code identical with that in Appendix A to form synits of

the first order, can be made according to the rigid order as follows:

(N) - 1x

1x 1x
(A) - 2x

2x 2x

32 2x 32
(v) - Ax

Ax Ax

31 Ax

33 Ax 33
(P) - 5x
( ) - 62
#3 61 63

61 1h 53

52 61 la
(C) - 7x
(S) - 8x
(35) - 9x

*
Figure 10 Formation of Synits of First Order

 

*
Each pair of symbols, such as 1X or AK is to be read as a two digit
decimal number whose least significant digit is unspecified.

-37-

Thus, for example, number Signits of the first order are of the

form 61 63, i.e. numerical number followed by a number-associate noun; etc.

Synits of first order are deve10ped into the make-up of synits of
higher orders. The following is a list of the synits of second and third
orders, derived from synits of first and second orders respectively. Assume
double brackets for second-order synit and triple brackets for third-

order, i.e. noun synit of second order is represented by ((N)), and third

order by (((N))).

((N)) - (N)
- (A) (N)

((V)) - (V)

((P)) - (P)
(P) (N)

((27)) - W) (N)
(Z!)

((0)) - (C)
((3)) - (S)
(($)) - ($)

Figure 11 Formation of Synits of Second Order

-38-

(((N))) - ((22)) ((N))

- ((21))

- ((N))
(((V))) - ((V))
(((P))) - ((P)) ((N))

- ((P))
(((C))) - ((0))
(((S))) - ((5))
((($))) - ((55))

Figure 12 Formation of Synits of Third Order

It can be seen that there are nine part-of-Speech classifications
(Appendix A), and the types of synits of the first order, derived directly
from these parts-of-Speech codes, is numbered at eight. The number of
different types of synits of second and third orders are further reduced
to seven and six respectively. The formation of synits is to set the
stage for the next step of the syntax process, that of examining sequen-
tial ordering of the synits. Therefore, the contents of synits of the
highest order must be so constituted that they are compatible with the
ensuing needs. For certain, each synit shall represent a distinct entity
whose properties are only known through the synit with which it is idenr
tified. An examination of the synits of the third order shows that the

conjunction, special term, and punctuation mark synits are identical in

-39-

content as with the respective synits of first and second orders. In

the case of conjunction and puntuation mark synits, their functions are
that of subordinating, coordinating, and ending a clause and there is no
meaning or concept relationship between these words and other types of
words. Special terms, as indicated, are words or groups of words whose
grammatical roles are quite complex and therefore should be isolated.

The remaining synits of third order are noun, verb, and preposition
synits. One may contemplate a further merging of these synits into
synits of higher order. However, a closer investigation shows that

this will defeat the purpose of synit formation. If the noun synit is
combined with the verb synit, then their combination will generalize the
conceptual or thought content of the new synit to such an extent that

the unique distinction of synit content is lost. For the same reason
combining the preposition synit with the verb synit leads to a similar
result. The remaining possibility is between noun and preposition synits,
but they are both formed from noun synits of second order, therefore the
significant relationship between these two synits are already processed
during the formation of third order synits. It is then concluded that
third order is the upper limit of forming synits. Further evidence will

be found in the sections on synit sequences.

The algorithms for formations of synits are now described. The
string of Chinese words ending with a punctuation mark is again represent-

ed by

-ho-

W W ... W ... Wfi, where W

l 2 k lst word of string

Wk = kth word of string
W? = a punctuation mark.

and abbreviations for the different fields of the information box also

conform with those previously designated.

Each word shall be preceded by a symbol indicating which type of
synit the word belongs to. The symbols for the different orders of

synits are arbitrarily assigned numerically as follows:

Table 5 Code Assignment for Synits I

(N) - 11 ((N)) - 12 (((N))) - 13
(A) - 21

(V) - 1+1 ((v)) - 1+2 (((V))) - 1+3
(P) - 51 ((P)) - 52 (((P))) - 53
(fl) - 61 ((7.2)) - 62

(c) - 73 ((c)) - 73 (((c))) - 73
(s) - 83 ((8)) - 83 (((s))) - 83
($) - 93 (($)) - 93 ((($))) - 93

NOtice the composition of words in all three orders of conjunction,
special term.and punctuation mark synits is identical, consequently

the code for each is the same.

-hl-

Each word which is an initial element of a synit is preceded by
one of the above symbols, which serves the same purpose as the left-hand
bracket. Other words belonging to the same synit also need preceding
symbols which are not necessarily identical to that of the inital element.
These symbols merely indicate that more words are included in the synit,

and are assigned as:

Table 6 Code Assignments for Synits II

Word interior to the noun - synit - l5
" adjective - synit - 25
" verb - synit - #5
" preposition — synit - 55
" number - synit - 65
" conjunction - synit - 75

Special term- synit - 85

Suppose the words which form a noun synit of the first order are W WhW

3 5

(i.e., p >.5). Then they are recognized by the machine as
11 INFO (W3) 15 INFO ("1.) 15 INFO (W5).

The symbols for the synits associated with each word, such as 11 and 15,

shall be abbreviated in general by SYB Wk.

As each word in the string is processed by the computer, the
machine must be provided with three pieces of information in order that

the apprOpriate choice of synit Sign preceding the word and the location

-h2-

where the word is to be stored can be made. These are (l) the symbol of
the current synit, (2) first location of the list of INFO(Wk)'S and their

preceding SYBW ’5 starting with W and (3) the location where the current
k

l)

INFO(Wk) and its associated SYBw must be stored. For this purpose, then,
k

three memory locations, to be named s, f, and z, are allocated to store

the reSpective information.

The procedures involved in forming the various synits follow iden-
tical patterns. Since the formations of the noun synits embrace more
possibilities than others, they are described in detail in the following
charts; and formation of other synits can use the same charts with the
only difference being that the part-of-Speech code and the synit symbol
preceding the word are changed wherever applicable. The algorithmic
charts for forming noun synit of first, second, and third order are

given in Figure 13, 1A and 15 reSpectively.

3d. Synit Sequences Involving (51/?

If the same or very similar structural devices are used to rep-
resent relations in both source and target languages, the transformation
of relation representations would present little difficulty. However,
Chinese and English do use different word orders to denote the same
grammatical relationships. Therefore, synthesis of Chinese sentences
into English requires not only a clear understanding of synit content,

but their sequential formations in a sentences as well.

-43-

 

 

2. Load Content of f into 3

 

 

 

 

 

 

 

Specified in 2

Load INFO (Wk)

 

 

 

 

 

 

Store 93 INFO ‘J\I§§l
(W ) starting
t location

2
P08 (wk) a 9x

 

 

No

 

 

 

 

 

 

 

Next Stage

 

 

 

POS (wk) 3 1x

No \

 

 

 

 

 

 

 

 

 

 

 

 

 

POS (Wk) must satisfy one
of the part-of-speech
Then decision is
made accordingly to go
to the particular synit
formation chart.

codes.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1. Store 11 in S

2. Store 11 INFO
(W ) at location
8 cified in l.

 

Yes
Content of S 3 11 YES
I No
Content of S i 61 Yea"
No
/
Store 15 INFO

(Wk) at location
Specified in z.

 

 

 

 

 

 

 

 

 

Store 65 INFO
(wk) at location
Specified in Z.

 

 

 

 

1. Adjust content of z to indicate storage
location of next word.
2. K + l -———% K.

 

 

Figure

13 Algorithmic Chart for Forming Noun Synit of First Order

 

-hh_

 

lo K=l

2. Load Contents of f into 1

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Load 3
I YBwk a. INFO 1
(W ) '
A Yes Sinsw 3 93
k
‘L No
7 Z No
Store 93 INFO SIBwk _. 11 |____,
(w ) starting
P Yes

 

at location
Specified in l

 

Content of S g 12

 

 

 

 

 

No

 

 

 

 

Next Stage

 

 

 

 

No

 

‘

 

Content of S i 62

i l

 

 

 

 

 

 

 

Content of S i 52 (£1 fl

 

Yes

 

SYBw must satisfy

one 8f the symbols
for synits of first
order, and apprOpriat
decision is made
according to SYBw

to go to the particul
synit formation chart

 

 

 

a)

 

1. Store 12 into S Store 65 INFO
2. Store 12 INFO (wk) at
(Wk) at location location

 

 

 

 

specified in z.

 

Store 15 INFO
(Wk) at

location
Specified in l

 

t

 

 

Store 55 INFO
(Wk) at

location
Specified in l

 

 

 

specified in I

 

location of next word

1

1. Adjust content 1 to indicate storage

 

 

 

 

2. K + l ————9 K
' I

Figure 1A Algorithmic Chart for Forming NOun Synit of Second Order

 

-h5-

 

lo K=l

2. Load content of f into £

 

 

i

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

SYBW must satisfy one of

the Symbols for synits of
second order. Then
decision is made accord-
ingly to go to the partic-
ular synit formation
chart

 

 

 

 

 

 

 

Load smwk & INFO (wk)
A/‘EEE. SYBW 2 93
k .
V No
‘3 NO
Store 93 INFO SYBw = 12
(W ) starting at k '
P Yes
location Specif- AL.
ied in £ ?
, Content of s e 13 Yes
‘I No
NeXt Stage I I Content of S = 53 35 \
No

 

 

 

l.
2.

 

(Wk) at location

Store 13 into S
Store 13 INFO

 

 

 

specified in z

 

 

 

 

Store 15 INFO

Specified in z

 

(Wk) at location (Wk) at location

 

Store 55 INFO

Specified in 2

 

 

 

 

 

 

 

it

 

 

1. Adjust content of Z to indicate storage
location of next word

_ 2. x.+ 1 ———————> K

 

 

 

Figure 15 Algorithmic Chart for Forming NOun Synit of Third Order

 

-Ng-

Synits partition a sentence into localized thoughtgroups, but to
weld these different groups into a complete expression requires further
investigation of relative sequences of these synits. Additionally, the
Special terms and the potential syntactic functions of individual words

also provide key information toward syntax analysis.

As indicated in Section 5 of Chapter I, a classification of part-
Of-Speech called Special terms is necessary due to the complexity of the
grammatical uses of these terms. Possibly the most frequently used word
among the Special terms is the word 97 , which is one of the chief
means for the detection of certain relationships among synits in a sen-

tence.

All synits now considered are of third order. ( [ ] is used as
equivalent to ((( ))).) The particular sequence of synits that is signi-

ficant is formed as follows:

[verb] [noun]: [ f5"? ] [noun]2

However, detection of such a sequence of synits is not sufficient
to guarantee a synthesis procedure into English. The reason is that not
all verb' synits will satisfy the given formation. Consequently, the

information box of the verb in the verb synit must be examined, and in

 

Presence is Optional in certain cases.

-h7-

particular the potential syntactic function part of the word. The in-
dicating Sign is thus found from the combination of its part-of-Speech
and its potential syntactic function. In case Of more than one verb in
the verb synit, it is assumed that the initial verb Should be examined,
and the assumption is made on the basis that the first element generates

the action or thought involved in the verb synit.

The verbs must also be separated into two categories. The first
category includes intransitive verbs and verb-prepositions, and the
second includes transitive verbs only. The intransitive verb- 9? or

the verb-preposition- (3’7 sequences are formed by the following order.

1. [verb ][noun]:[ ‘9’? ] [noun]2

intransitive

l

*
2. [verb-preposition] [noun]l [ \F? ] [noun]2

An example for each case is shown below:
1.E’_ni_£][3e@1[éfy ][_/_\_1
To live in. ‘ U.S. PeOple
Translation: Pe0ple who Eé live i2_U.S.

2.[£][\$7#§2fl1[é’71[:@afi731
To be in 'Anartic continent 'One classifier place

 

 

Translation: One clfisfiifier place which is in Anarctic Continent

 

 

 

*
Presence is Optional.

-h8-

The synthesis procedures take the second noun synit, i.e. [noun]2 and
places it in front of the predicate clause of [verb] [noun]l and the

word g’y is deleted.

To make the English translation correct, the conjunctive pronoun
"which", "who", or "where” must be inserted following the [noun]2. The
choice should be made to agree with the substantive in [noun]2, but this
is difficult if not impossible. We shall merely indicate that 8. [NH]

term is to be inserted to stand for either of "which", "who", or “where".

The #7 in the above case has only one possible function because
of the intransitivity of the action involved. If a transitive verb takes
the place of the intransitive verb in the same formation of synits then
the noun synit [noun]l whose presence was optional previously, is now a
necessary element to follow the verb. The usage of {347 here can have
two interpretations. (5/9 's function may be that of making the case
of the noun synit preceding it possessive. Then synthesis merely in-
volves a process of combining this noun synit as a modifier to the noun

synit following {34] 3 é’g is deleted but no order change is necessary.

On the other hand, the word Q47 may be used to effect a change

similar to the previous intransitive verb case. Examples:

1. [ stars 1 [3:61] [1’34] 1 [232% 1
' To visit HIE: Scholar(s)

 

Translation: Scholar(s)j[wH] t¢ visit U.S.

 

 

-hg-

2.[%_§L_][—$si][é’-7 J[£]
To teach ‘ English ' Porsongs)

Translation: Persongs) [WH] Z2 teach English.

 

It is important that the two different functions of #9 must be dis-
tinctly recognized. The prOper choice again goes back to the stage of

meaning determination where human intervention is necessary.

If by now {3’9 still remains in the passage without satisfying any
of the formations thus far considered then its presence Should be located
either following a noun synit or a verb synit. In the [noun]- éfiy case,
the result is to make the noun synit a possessive case, i.e. [noun]'s,
and in the verb case the verb is made into past tense. 9’? is deleted

after the change is completed.

The algorithmic charts for processing the synit sequences involving

{247 are shown in Figure 16 and 17.

3e. Re-formation of Synit of Third Order

It is evident that the synit order may be altered after process-
ing all synit sequences involving ’3‘.) . The possibility arises from
the fact that some of the synits which were separated, now form adjacent
synits which are so related that they can be combined into a Single synit

according to the rules shown in Section 3c (Figure 15). HOwever, these

-50-

 

 

 

 

 

 

 

K + l—)K \ Load INFO (wk)

POS (Wk) 3 9x —X§§)+ Next Stage

N° Pos<w>18u<t<7>

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Does the Synits in the
No neighborhood of 87
form sequence

[v] [N11 [947] [N]; r

ALures

Rearrange order of
Synits into
[N]2 [WH] [v] [N]l

J,

delete g‘]

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 16 Algorithmic Chart for Translating A Chinese Phrase whose
Sequence of Synits is [V] [N11 [ 97 1 [N12

 

-x-
The use Of preceding Signs to indicate types of synits for words be-
domes significant here. The detection of the particular sequence of

synits, such as [V] [N] [99] [N] above, is to be made based on
those signs. Subsequen syntax algogithms dealing with synit sequences

also employ this Scheme.

-51-

 

 

 

 

A.

Load INFO (wk) I

4.

PCS (Wk) .3.- 9x |_x__)es Next Stage

\IINO

K+l—>K |<—N°—'POS (wk)%8l+ (6'9)

#JkYes

K - l -———fi> K

J

svsw lNoun Synit Yes

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

k
No
NO 1
-- K+2—)K le— STE" -Verb Synit
k
Yes V
Make the Verb Make the Noun
Synit into Past Synit into
Tense Possessive Case
delete (91:7
]

 

 

Figure 17 Algorithmic Chart for Processing Synit Sequences Formed by
W Preceded by a Noun or Verb Synit

752-

 

l- K = l
2. Load Content of f into .8

l

-—)| Load SYBWk and INFO (Wk)

EL

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Yes
——-—’ smwkij 93
- SYBw must satisfy one of
iNO the k symbols for synits
- 7 No of third order. Then de-
Store 93 INFO smw = 13 ,
(w ) starting k ' cision is made accordingly
P Yes - t t th arti 111
at location 0 80 0 e P C ar
SPGCH'ied by 3 s it formation chart.
Content of s 3 13 Yes yn
INC
No
1. Store 13 into S Store 55 INFO Store 15 INFO
2. Store 13 INFO (Wk) at location (Wk) at location
(Wk) at location Specified by z Specified by l
Specified by l .

l l

1. Adjust content of z to indicate storage
location of next word.

‘ I

Figure 18 Algorithmic Chart for fie-formation of Noun
Synit of Third Order

 

 

 

 

 

-53...

established rules were based on information of synits of second order.
Otherwise, the procedures follow the same format. In Figure 18 the al-
gorithmic chart is given with.modifications made to deal with synits of
third order only. The formation of the noun synit is again considered

to exemplify the general procedure.

3f. Other Synit Sequences

It is obviously not possible to enumerate all the syntactic
structures in the Chinese language which are different from English.
EVen if a certain sequence of synits is detected that effects an order
change for synthesis into English, one may well find a similar sequence,
in which the words used are different, which does not need such a change.
It can be seen that in the synit sequences discussed in the preceding
section the order change is made based on two factors; namely the pres-
sence of the word Efiy and agreement on the syntactic function of the
verb. Therefore, it is extremely difficult to establish general rules
indicating only the type of synits in a sequence without any regard to

Specific words.

In view of the irregularities in word structures in Chinese only
two Significant synit sequences are found to be meaningful when their
orders are changed. The first deals with a sequence initiated by a pre-
position synit. The second involves a statement of comparison. In both

cases the syntactic function of certain Specific words needs to be examined.

-5h-

The formation of the prepositional sequence is
*
[preposition] [verb] [noun]

Some illustrations are first given.

l.[\:]1j’;] it] [fl]

Th you To bid Farewell

 

Translation: To bid farewelltg you.

 

2.[fifi][§:][’3’11
For mg. To buy Past-tenser

 

 

 

 

Translation: Tb buy past tenSer for me. (or bought)
I -—' 4

 

3. [ @3221 [i1

 

” From home To come

 

 

Translation: To come from home.

 

The synthesis involves rearranging these synits into [verb] [noun]?
[preposition]. The flow chart for automatic processing is given in
Figure 19. The preposition used for this particular function should

be provided with Sign in its potential syntactic function field. A list
of some of the prepositions more frequently used in the prepositional

sequences is shown on the next page.

 

*
Presence is optional.

-55-

Table 7 A 1.181: Of Frequently Used Prepositions
in Prepositional Synits
[ ['5] ] [verb] [noun]
] [verb] [noun]
[ [it ] [verb] [noun]
[ Sui ] [verb] [noun]
[ ‘3] ] [verb] [noun]
[ ”[5; ] [verb] [noun]

[ [fl ] [verb] [noun]

The statement of comparison usually uses certain comparative signs.
The common ones are $3 and 3%: . Thus "tall" ( (137 ), "taller“ and
n n a; 4; g 'é,’ -
tallest in Chinese are “7] , f ‘37 , and 3:1 ‘67 . However, when a com
parison is made in which one member is expressive of inequality, the word
[:L: is used. [:13 is placed before the member expressive of inequality

and the measure involved after. Examples:
1° 33 fit is s
_I_ _I_I_e_ Tall

Translation: 2_[_ taller than he.

or _I_ am taller than h_e_.

2'“i ii £13 5.4%

That Classifier Tree Tall Three Times

Translation: Three times taller than that classifier tree.

-56-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

_ifi:3 SYBW' & INFO I

 

 

 

 

 

 

 

 

 

 

 

 

 

SIBwk ; 93 IZE§—;+Next Stage
NO _ -
K+l-)K eflsmwk ; 53
I Iles
Does the order of
NO next two synits

 

 

 

 

satisfy [v] [N] 7

 

I'Yes

 

Rearrange
[P] [v] [N] into

 

 

[v] [N] [P]

 

Figure 19 Algorithmic Chart for Translating a Chinese Phrase
whose Synit Sequence is [P] [V] [N].

-57..

The Change of word order that is necessary to give correct trans-
lation into English involves the placement of the adjective of comparison
in its comparative case and the quantity, if Specified, before the member
which is compared. In the case as illustrated in unple l, the word {$7
(tall), or in general the adjective of comparison, is made comparative
(i.e.taller than) and placed before the element being compared, fig (he).
In a.more complicated pattern as Shown in Example 2, the additional part
I of the phrase, i.e. 3. 1% (three times), is placed before the comparative
adjective ”taller than". The word bésis deleted after word order is

changed.

The sequence of synits is not directly reSponsible for recognition
Of statement of comparison. The key to detection comes from.the syntactic
functions of the words H: and the adjective of comparison. But Al; is
a Special term and therefore, forms a synit by itself. The remaining
words following at in the statement of comparison may form one noun synit.
However, no ambiguity Should arise since the syntactic function in adjec-
tive of comparison is first detected and the prOper word order change is
then made within the synit. The algorithmic charts for processing the

statements of comparison are Shown in Figures 20 and 21.

3g. Output of English Equivalent

The final stage of translation, that of the output of the English

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

K = 1
Load INFO (wk) I‘
2 Yes Q
POS (Wk) a 9X Next Stage

‘ No
Agfi * YES
FUN’(Wk) = 21 ($3 )I——————

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

N
K + ln——>KIkAELI FUN’(Wk)'; 22 (ifif)
, f A Yes 7
f -
K+l—)K I K+l—)l( I
Load INFO (Wk) Load INFO (Wk) I

?
-—-———3EL——IPOS (wk) 2 2x POS (wk).§f 2x IJEL—-
- ‘ Yes IYeS

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Adjective Wk made Adjective Wk made
Superlative Comparative
Idelete INFO (wk_l) delete INFO ("h-l)

 

 

 

 

 

 

 

 

 

Figure 20 Algorithmic Chart for Processing Comparative and
Superlative Words in Chinese

 

* _

For convenience, the syntactic functions of i and 3% in their roles
Of comparative and superlative cases are arbitrarily assigned as 21 and
22 reSpectively.

-59..

 

 

 

 

 

 

 

 

 

 

 

 

J Load INFO (wk)

,L

 

7
PCS (wk) . 9x

 

 

 

INo

 

POS (wk) .1. 83 (kt)

 

.IYeS

 

K+l——->K

 

 

 

 

 

Load INFO (wk)

 

 

 

 

 

 

210—4

 

Yes

POS (wk) 2: 1x

 

 

 

 

Rearrange word order 0

ts -

comparison - number -
number - associated

Noun's - Adj. of

 

 

 

 

 

 

l

 

K+1—>K

L

Load mo (wk)

J

'2
P08 (wk) = 6x

 

 

 

L753:

I__N_<L.

 

K+l—>K

 

 

I:

I Load INFO (wk)

 

 

\[r

Y” POW (wk) 2 63

 

ALE)

 

order of bh - Noun's -
Adj. of comparison to

 

Rearrange original word

"Adj. of Comp. (compara-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Noun to 7 I tive case) - Noun's
3‘ \L L
2 4(- [e
/ NO FUN (wk) a: 23- delete Hz
J

 

 

 

 

 

Z;

Figure 21

r-Nmber-Ass. Noun- Adj. of Canp. (Comparative Case) - Noun's.

Algorithmic Chart for Processing

Statement of Comparison

 

k

* The syntactic function of the adjective of comparison is arbitrarily

assigned as 23.

s60-

counterpart of the Chinese passage, is a matching process since the
locations Of the English equivalents Of the words can be found from
their information boxes. Presumably the English words are stored in

the memory, and it is only necessary to go to the particular location
and bring out the word for an eventual print out. The other fields in
the information box, the part-Of-Speech and potential syntactic function,

no longer serve any purpose and can be deleted.

h. Conclusion

Although machine results are not available at this time, the
syntax routines have been tested repeatedly by hand by following the
algorithmic charts Shown in this chapter. Some simulated results worked
out by hand are shown in Appendix 6. The passages translated are selected
so as to represent the various cases Of grammatical patterns in the

Chinese language.

Emphasis has been_previously'made on the fact that the translating
system.thus developed is designed to translate fairly elementary Chinese
texts, as exemplified by the reference books used. There is no doubt
that this system will encounter many difficulties when tried upon any
text of'more complex nature, both in grammar and word usage. However,
it is believed that the schemes developed, particularly the formation of
synits, are of considerable value toward future work of mechanical trans-

lation from Chinese to English.

APPENDIX 1

THE RADICAL LIST

-61-

 

 

 

 

 

taste this asi$és$fitsmxsss
Duals; alvixlasmihaismsmssfleraahsm
Miriamsarl ailisihriahh
whamthkkwhs artisans..."
:[panuilatmtfl aw%asa¥atmi
e+gF attf warrantsatxs
nrrt- is hashttara
AC; a hathhflass
thitmtwt a. assistants.
rosette/ass i “Mamastrawastntn
ms .3 s assmma: mamas;

 

 

 

 

 

 

APPENDIX 2

A LIST OF luOO COMMONLY USED
, CHINESE CHARACTERS

-63-

-54-

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Feather—malarial]? Ease.safestvsixstsrtar$es
fisdaaaaaasshhhhswaaafisiii
Tslainsigfiffitgkttttkttttfi
swirl hiafiiawaaiirtasfipnwssrsnatal
varietisawfishlasatsfialwsmheartachewh Em male
ants/hates Estartfiwxvarxeseihmaswghis
airmaiw a ten Rhonaqéllxtuaae/Sls smflaewltaers
walls—aha WEWEWWmEfirahfi/l lass/e Zahara
ahattahahiatm5§raE++wtttm
hflflflwflfiififlwflanflmtmmtawaniiwmuwsnfi
EdutainagsitrfiimhrmNamath
seesaw? .tiitttinftf has mm as stamina
heahemhmgeehfihfihwahhtaoofi
mafiathetahehaasfitmgiwaaat
_siivlEhreisihit9xflestmmhfi
_i_tilitsatisisiinxztistet

 

 

 

 

 

as it if 2%:

 

 

 

 

 

 

 

 

 

 

 

 

 

at

first

has

the, ashes e. as. as assertssratisinissrss as stain. H
Nhhfimtaathinflm Balersswnasamiéarasas
aaaaasdaaaahtaaraafisaith?
htthhfithhtthhmaFEEhfianiE
s aaahhaathaaaahehéthaastfia
asihmhhmoaaafitiflhaflePatio titan.
hawmediatorsmantrashe[hissifiseEasaawttFMMs/Maafimfir
mm ”We—rm. me M aw two .3 a]... Re R Aims to a; as .m lbw. E as are Mm Nb aw mw kw
_Wfl$%fiflfi_%_wttk_m l/wmmfimmfifirshdmfisfimmfi
Firmwfinifisfiifififiafifia?szRN
rttifihiflfiéwFFfiWfiTFfitfiinTI
EaRnfihaéhifiééfifiht$$53WMgm
.Tm..fl$$§.r®.me®ifiaxfl #tTfififivgxmfiamfifislILM [wltfimmmw t

 

 

 

-66..

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

_m mtearnhandlebarssaitxarssmstamafieaasiRestraint
simsaishaaiaseaasaaaashae
Elamamumeasasaaarvfimmtmfinwwheafimream.
_atrfiwflmtimtwfiafiFarhhmatam
atawrhmaaanafintahfiaim harm with m travails...
fiifim{Thaimavniszasaecstasaharam
amatirllslmwaimmskatinsuxsaawassamamifisafi—mlemmaonfiamfi
tom—m ihflwaahohifis/m (siesta [assessafim ll
tesafiflmshfimsatarsfitaaawtt
loam madam e Ermaimimmfi ..Eemiaseaaim Sightseer
rim Trim mt snowflursms ETasislstaS M Kim fihlmlhifi
Whfiflwmmmfiglfflwitems/meanwgfisnw/Eiaami
meter assesses; this? This intimhfa m
fifissasérklgsswmltflhttemahassaaiws.
asistsélrrtaiatafiafisrrshattersih
Tashaislethatasaaauaisssstxafitarseta

 

Q“

q‘u-

“\u-
a A”?

)‘I

\«e ow- \w “\w our our m.» mm
' 2“: 356 I“ \-/~ 4»

fl "3'
G I"

,VH'

M“- om- MW

(2% W

" NE?

3.

CW EIi‘fi‘: {a

I]

21”:

on, e\\\- 0
75°

as:

\w “\u»

we

“(...

—i— ‘-\
one u
an

\W

.35

fl

11hr

are

our
ash

 

 

 

o\\\- cm at» n\«- ON—

['0
ow
is?”

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

afar—2%: its. this FR #131173183’53 It
shiatfssagfnwhgéhmfi
arraisaptaitttfifiht
sfisarh seat- Fri—922‘s:
s7t333159a, or iiftfizgfikrfiéé
sZEFfiaIars—wa-Eefiafiias
ssh—nathispssaaa at?
harassirtaaattfis
a treatise Fi9¥IE5§195§i
at] ”s? tripart‘ with ”aa-
éE‘é‘fBE—iiii—la’ié’é has??? «it—503
at? t was 1 at a 185196316}
iiswfiaafzaflafs @207‘
she—E’oirtssiEiEh-w—Mtfii
at? that a EHGR‘EEM
it’s-sari Pfifiiiégfiifiv
firtafiafiahahsgza
f‘a’amzfie’t raififigfitmi
: trail—63:11: at Ira-13353210
isofi’ismiiflflifgifi‘fi’fififi
Elsi}? sat—175:1; 1781:; 13213531}
trawliaaagfig’rm
rattaifl’f atmweayofii
gggfilzglfimgfigg 5119020022
gggfsfii’f Nimii‘faiéhzlu

 

 

APPENDIX 3

A SHORT LIST OF CHINESE I CHARACTERS COMPILED
ACCORDING TO THE RADICAL SYSTEd

-69-

The word is listed on the right Side of the column. Its associated

radical is indicated at the upper left corner of the first word of its list.

The numerical equivalent for each word is arbitrarily assigned and appears

«-
at the left of the word.

 

 

 

 

 

9(-
3'Using the CDC l60-A Computer, it is conceived that dictionary information
Hence, starting at loca-

 

for each word takes ten octal memory positions.

tion 1000, the digital equivalents are accordingly assigned.

 

 

1:00 ... 1110 i 1420 5" 1630 (a 2040 i 2250 A) 2:60 A.
2: s a] a s, t it, rt t
1010 J: 1230 4'3] :40 t 0 1650 4; 2060 % :70 ‘I‘ 2510 3t.
1030 T 1140 [it 1450 I” 1660 J;- :10 it 2300 E. 2:0 ’3']
1740 7"" 1150 [P9 Z460 E 1670 7’5“ 2.100 11° :10 {It} :10 “I
[:50 :[D 1260 {K 1470 + 1700 a 2110 t; :20 AF 2540 IE.
1260 Z 1270 1i 1500 f 1710 F“? 2110 S 2330 $- 2550 f]:
11070 t2 {500 /\ 1510 "I“ 320 ’E‘ 230 if? 2:540 {E 2560 ”[313:
Moo 7 1210 £3 I152s ‘33 1730 E 2140 g, 2350 [a 2570 #2
1.110 $ 1310 It 1:30 E] 1740 '3 :1‘150 \i'?’ 2360 [FA :00 at
I, 20 {E 1:30 Ii 1:10 7?? :50 [a :60 ‘8 hi?» [‘3' 2610 5i
1130 L‘/\ 1340 7‘? 1550 1 2000 ’2: 2170 \ifi 2400 'I'Jé 2620 at
1140 ’[E 350 '3‘. 1560 R 1760 3 t 2200 E 2410 'I'é 2630 "gt
a a BI {is E ii .... f: is i
”6° 1F 1370 ”k 1600 g :10 #I‘ 2220 ‘ 2430 [3‘ :50 1%
1170 If]: 1300 )1? l6/o P‘ :10 g 2230 52:7 24140 :1 2:0 7;
120° ‘9‘ 1410 3'] 1620 “Z 2030 k :40 'J ‘ 2450 '[i’ 2670 35‘

 

 

2700

2710

1730

27%

2500

2'15” 0

1760

2770

3°10

3°20

3000

3a30

3°40

3°50

3°60

3°70

3100

3u0

3110

£930

 

31110
61:.

3150

3160

3170
5

3200

3210

.‘
3110

3230

31% ‘

3250

3260
k my

3270
3300
“K
3310
3310
II-

3330
33%
3350
3360
3370

3400

 

lywo
34.20
3430
l?
3‘140
3450
7°.
3460
3670
3500
3520
3530
3540
3550
3560
3570
3600
3610
3610
1:):
3630
3640

3650

3660

*anfihimmgmxggg

3m

sassmhgtgyi

 

7t

’
—
a

I

-70-

3750
3b
3760
”COO
4010
402.0
é 3
9e»

M0

4M5
4-070
4100
4M0
4n0
4130

4140

Ex»
or

2»
I423

assesses

153-

5515"

EzEnfl‘

I

3* has

at

1,?

 

emssmsts-

‘HSO

4160

4170

4mm

4UUO

4110

42.30

4a»

4250

4260

4270

4300

4310
4320

4330

434.

4.350

I

as

\i; chi- M. m». o\w
16's to; 1“?

«ID

3* e

the

II»-

 

sss 1"

4430
[“3

’1‘4‘1‘0
4450
4460
4470
4500
45/0

4520

4530
it
45%
\i7
#550
4560

4570

46/0
4620
4630
4640
4650
13k

#660

4670

starsmbssssasasagzsgges

as

{P-

 

as as [as 5,»

J}

‘?
\tq

4700
4WD
3770
W20
3510
4720
4730
4740

4‘15 0

4760

JO

Ll 1'1 v A

 

APPENDH 1+

CIASSH‘ICATION OF PART-OF-SPEEH OF THE CHINESE WORD

-72-

-x-
The numerical code for each classification is assigned arbitrarily.

Sample Words of the

 

figu§§_(lX) Classification
Nouns <11) m. s . si
Pronouns (12) {I}: ' flffi
Pr0per Nouns (13) a $
Time-Associated Nouns (11+) 3., g, 31;}
Adjectival Pronouns (15) 1%., ’F
Conjunctive Pronoun (16) [NH] / ;.¢. ubicA,on, 06ers

Adjectives (2X)

 

Adjectives (21) 3; 7g: , 33:31

Adjectives of Comparison (22) k , é?
Adverbs (3X)

Adverbs (31) 37, fl , R

Adverbs of Comparison (32) k , 1%.

Adverbs of Condition (33) >3. ’37, @534
we 0x)

Transitive Verbs (1+1) at,

Intransitive Verbs (1+2) 5,2} is , $3.

Auxiliary Verbs (1+3) ‘ @‘I Q

Verb-Prepositions (Ms) E, ‘%

Prepositions (5X)
Prepositions ’ (51) fig
Time-Related Prepositions (52) J: ' 11 ’ if} 3541.

 

The two-digit number with X as second number denotes the general
category, such as fix for verb; while a numerical second number denotes

the particular sub-field of part-of-speech, such as #2 for intransitive
verb.

-73..

SMple Words of the

 

Numbers (6X) Classification
Numerical Numbers (61) ..., .31 ) +
Pronominal NumberS‘(62) .— {13 ’ gr 3,

Number-Associated Nouns (Classifiers) (63) 1.3] I if, )1

Conjunctions (7x)

 

Introductory‘(Subor.) Conjunctions (71) #9 .fi . {g i ’ “I M
Coordinate Conjunctions (72) *9’ 51‘
Special Terms (8X)
Special Elements I (81) E ’ %
Special Elements II‘(82) {E a
1.1:. ‘ (831
947 (81+)
Past Tensers (85) '3 , 1%.

Punctuation Marks (9X)

kriOd "' 0
coma: " ,
Colon - 3

Semi-Colon - :

Brackets - ( ), [ ]
Quotation Marks 4- "
mclamation Mark - '2

APPENDIX 5

LOOK-UP PROGRAM AND SAMPLE RESULTS
USING CDC 16o-A cmurm

-7h_

-75..

Master Routine

Function Execution

Code Address gpgrand
JPR , Subroutine I
LDD 75

SEM 1000
SBN 77
ZJF 03
AOB 03
NZB 10
JPR Subroutine I
LDD 75

find 0105
SBN 77
NZF 02
Go to Next Stage
JPR Subroutine I
LDD 75

END! 0106
SEN 77
ELF 05
LDN 03
RAM. 0105
NZF 33

IUD! 0105

Comments

Input dictionary in-
formation

First Storage location
= 1000 -. . :

End of list check

Input word to be looked
“P

Temporary Storage 0105
End of list check

Input following word

for idiom search

Tumporary Storage 0106
End of list check

-76-

Function Execution
Code Address gErand Comments

STF 02

L114 *l-l-I-

8m 0106 Check for first idiom
in dictionary

ZIP 12

LDN 03

RAM 0105

STF 02

L114 «m

SM 0106 Check for second idiom
in dictionary

NZF 12

LCN 02 Output idiom information

STD 77

AOM 0105

JPR Subroutine II

AOD 77

NZB 05

2.13 5h

LDN #93 Output word information

RAM 0105

JPR Subroutine II

AQM 0105

JPR subroutine II

L114 0106

STM 0105

SEN 77

NZB 63

GotoNextStage

Subroutine I

 

Function mecut ion
Code Address gpgrand

J'PI 01
*X-H

EXC ‘ J+102
IMP 03

PJF 02

010].

L114 0101
LS3

SCM 0102
LS3

SCM 0103
LS3

SCM 0101+
STD 1'75

PCB 211»

NIB 25

subroutine II

 

Function mecution

Code Address garand
JFI 01
a-x-x-x-
LIM‘ 0105
SIF 02
m H-x-x-
3m 76
EXC 111011
LDD 76

LPG 7700

Caments

This routine inputs
11 octal digits

and stores in
temporary location 0075

Oments

This routine prints out
11 octal digits

Stored in location
specified by the content
of 0105

-78-

Function Ekecution
Code Address %rand Canments

LS6

OTA

LDD 76

LPN 77

OTA

PJB 22

MB 23

For the given Chinese passage:
7):: Ekaiiiififl SHAAAABHA 6‘61]: .
Input to Machine:
2170 10110 3550 113110 11320 1230 2150 3030 2000 3100 A670 2750
12110 3660 3110 A710 0077

Output:
00 12 21:20
00 31 3h10
00 111 2700
00 15 6110
00 63 61:00
00 12 5020
11 M1 0150
12 82 7030
13 11 6010
00 1+2 6120

APPENDIX 6

SIMULATED TRANSLATION OF CHINESE PASSAGE
INTO ENGLISH

-79..

-80-

The following three passages in Chinese are to be translated into
English. The correct English translation is first presented beneath
each passage. The translation using the Syntax System.deve10ped in
Chapter 2 are then carried out in ensuing sections. Each string of words
ending with a punctuation mark is first written, followed by the word-
for-word translation. Details are stressed on processing the significant
syntactic relationships existing in the passages. The trivial parts of

the translating process are not indicated.

I’ fii‘m‘oiafiné‘mtinegam Q
’5 :aiiéhrara‘y k sci/segue.
it Masai E saw, “magmas
a #1112 waft: £9.71”; 531431;."
Translation:
I do not know when does this semester end, also do not kDOW‘when

does our final exam.begin. I only know that the teacher already said,

"Before final examination begins, we must review well the old lessons.”
IL i3rfi‘aét'3'1441‘ifayfl‘.%zd, detail.

Translation:

Three years ago I went to your home to bid farewell to you, now

I returned.

-81-

III“ ‘éffit>§éb%fii, 5571<%F@4;»11;%fi
\fiyfitfiféfi-{m’flfi’ fi‘rffiflifi; \gi’: A}?!
+Etss,ssse$sea .tfi
Items”? swam, excesses
Lt%1&3§7h%+-%.P r’rtxtaiiiipésé Mk.
Translation:

When the weather is very cold, even sea water can freeze. But at
one place in the Anarctic Continent, there is a dead sea; when the
temperature reaches sixty degrees below zero, this sea does not freeze.
The scientists who went there to explore for experimental results, found
it contained salt eleven times more than ordinary sea water. Therefore
it does not freeze easily.

I-lfi 1 as; La messages

 

 

I do not not to know this classifier semester -§t_ what time to end
Deletion of word:
Delete 1g) (classifier) after 1: (this)
Formation of Synits of first order:

(1) (do not, not to know) (this semester) (at) (what) (time)

 

 

(to end)
Formation of Synits of third order:

[I] [do not, not to know] [this semester] [gt] [what] [time]

 

[to end]
Translation:

I'do no t td'know this semester §t_what time to/end
do not ' ’

-82-

2's 1 115.11.111st 1.212.911; #1113

Also do not, not to know we 97 final exam.at wh att time o

 

 

 

begin
Formation of Synits:

[Also do not,not to know] [we] [ __E/__f] 1 [final exam] [at]

 

[w___hat] [time] [to begin]
Sequence of synits: .

[we] [ [9’7 ] -—) [we] in possessive case—)[our]
Reformation of Synits:

[Also do not,not to know] [our final exam] [at] [what]

 

[time] [to begin]

Translation:

 

Also WM know our final exam at what time ty begin.
do not

its 321*- itj LL?- i ii

‘I onIy to know teacher already to say past-tenser
Processing past-tense:
M past-tenser ——5 said
Formation of Synits:
[I] [only to know] [teacher] [already said]

Translation:

 

I onIy tp’know teacher already said
7*

-83-

1“ i k 7% E. s w. 4-5.,
to be at final examination to begin before

 

Frame Construction:
to be at . . . before -——> before
Formation of Synits

[before final examination] [to begin]

The sequence of synit above is [P] [v], but the syntactic function
of the word "before” as found by frame construction will not be so assign-
ed to effect a change in the manner discussed in Secion 3f. of Chapter 2.
Otherwise translation is incorrect.

Translation:

before final examination to/ begins
7

Ffifltké a as mg a

213 must with old __ér’§7 lesson[s) to review we—ll

 

Deletion of word:
9/] after g , an adjective
Formation of Synit:
[E] [must] [with old lesson(s)] [to review} wellJ
Sequence of.Synit:‘ I -
[P] [v] [N] (Optional) to [v] [N] (Optional) [P]
[to review well] [with old lesson(s)]
Translation:

[113 must to review well with old lesson(s).
/ /

—84-

ILL: 14: 11111.1 xiii-ems. 21
t_ you

Three year(s) ago I go to,to your home obid farewell

Formation of Synits:

[three year(s) ago I] [go to,to] [your home] [to y__]
[to bid] [farewell]

Sequence of Synits:
[P] [v] [N] to [v] [N] [P]

[to you] [to bid] [farewell] [to bid] [farewell] [to you]

Translation:

Three year(s) ago I'go aggto your home to bid farewell tg_y22,
went

2.

 

 

fififiil
Now to return ‘}

 

Processing Past—Tenser

to return 3 -———9>returned
Synit Formation:

[N31] [I] [returned]
Translation:

M I returned

-85-

x k .
s a a i. a; a? 1111
to be at weather very cold a. time

III. 1.

 

Frame Construction:
to be at time ——-) when

Deletion of Word:
Delete word 67 after )Q‘

Formation of Synits:

, as adjective

[When] [weather] [very cold]

Translation:

When weather -is- verv cold.
————-I—

 

 

-I-—
2, _m as £1 5? 91$ 1
Sea water even can to eeze

 

 

Formation of Synits:

[Sea water] [even can to freeze]

Translation:

Sea water even can to freeze.
/

\

3°4flifi 331:]: .511 3’2: iii] m1
Anarctic continent fig one classifier place

Bu to be in .

 

Formation of Synits:
[guy [to be in] [Anarctic continent]_[_§_’L]

[one classifier place]

Sequence of Synits:
[verb-preposition] [noun]l ['93 ] [noun]2 --€> [noun]2

[NH] [verb preposition] [noun]l

[nnp nlggcifipr nlane] [WV] [to be in] [Anarctic continent]

-86-

Translation:

But one classifier pIace [VH1 to in Anarctic continent.
/7' which is in

 

L“? :Ji ii

There-to be one classifier dead sea

 

 

Formation of Synits:

[thereto/be] [one clasdifier dead sea]

 

 

Translation:

 

There tq/be one classifier dead sea.
/'

"i5

5' it. it. eli_ 11:: 5- new.

to be at weather col.d tg_ zero below sixty degree(s) éfiL time

 

 

 

Frame Construction:
to be at ... time-——> when

Formation of Synits:

 

[when] [weather] [cold] [to zero] [below sixty degree(s)]

[6191‘

Translation:

 

 

When weather -is- cold to zero below sixty degrees fl .

6.

Ft-

t‘r set: 757 s asst

This classifier sea also do not,not can to freeze.

 

Deletion of Word:

xi

Delete Jf (classifier) aiterfs (this)

-87-

Formation of Synits:

 

 

[this sea] [also do notjnot can to freeze]
Translation:

This sea also do gptjnot can tg’freeze
" not 7

7° 4: :12”: as 352 33' ““1 3‘; 56.1.2":

to go to,to to explore 97 science eXpert(s) to experiment

*Si

result-

 

 

Formation of Synits:

[to go tojto to explore] [6’9] [science eXpert(s)]

 

 

 

 

[to experiment] [result]

 

Sequence of Synits:

[verb—preposition] [64]] [noun]—-)[noun] [WH] [verb-prep.]

 

[science exPert(s)] [WH] [to go tojto to explore]

Translation:

 

 

 

Science eXpert(s).LWH+;§g=gpateth to explore to experiment
who went
ESEEEE
k \gfi A I E _ s).
8‘M2 is. 32$ 1:55» the
To find it _t_g contain . salt to be e;- ordinary sea

 

 

zia § +- 4%-
vats} mare eleven tifiés

 

Formation of Synits:

[to find] [1.3] [to contain] [ [9/7] [S_a_l_§] [to be] [i]

 

[ordinary sea water] [more] [eleven times]

 

 

 

-88-

Sequences of Synits:
l. [verb] [[5]? ] —-> verb-past tenser
[to contain] [ 6’7 ]——)contained
2. Statement of comparison:
[eleven times] [more than] [ordinary sea water]
Translation:

,To—Sind’it contained salt to be eleven times more than
Found

ordinary sea watero

 

J

i .
a

\|

9° ff-r le \‘E’ 1"

Therefore 'EE‘ do notjnot easily to freeze

lg:

 

Formation of Synit:

 

[Therefore] [it] [do notjnot] easily to freeze]

 

Translation:

Therefore it ‘dg=gg§z§§§_ easily ‘t6'freeze.
does not ’

 

REFERENCES

Current Research and Develognent in Scientific Documentation No. 11.

 

Chapter 3, "Mechanical Translation", pp. 190-228, National Science
Foundation, November, 1962.

"Acceneral Canpiler Capable of learning", Richard Fairbanks Arnold,
A Thesis, Master of Science, Michigan State University, 1958.

Shau Wing Chan, "Chinese Reader for Beginners", Stanford University
Press, Stanford University, California, October, l9h3.

men Ren Chao, "Character Text for Mandarin Primer", Harvard University
Press, Cambridge, Massachusetts, 1951+.

Tu Hsui-Chih ( h ‘2? in ). China Weekly, pp. 10-12, Marksman,
July, 1962.

Yehoshua Bar-Hillel, "Machine Translation of languages", Edited by
William N. Locke and A. Donald Booth, pp. 183-193, The Technology
Press of the Massachusetts Institute of Technology and John Wiley
and Sons, Inc., New York, October, 1957.

Paul L. Garvin, Proceedings of the National Syntposim on Machine
Translation", E. P. Edmndson, Editor, pp. 286-292, Prentice-Hall,
Inc. , Englewood Cliffs, New Jersey, 1961.

Anthony G. Oettinger, "Automatic Language Translation", Harvard
University Press, Cambridge, Massachusetts, 1960.

-89-

ROOM USE ONLY

 

 

IES

"‘fi@[i[ujui[u[[[[]][]’][[][]"