Transformational Dec~npos£t~on z 
A Simple Description of an 
Algorithm for Transformational ' 
Analysis of English Sentences~ 
Danuta Hi~ 
AravlndK. Joshl 
• University of Pennsylvanla 
ABSTRACT 
In this paper, we will present a ra~her simplified description o£ an 
algorithm for transformationalanalysls (decomposition) of English 
sentences, our purpose here is not to discuss the transformational 
theoryj the full details of the theoretical formulations of the algo- 
rlthmj or of the gran~ar. Ratherj we will present a set of examples 
of the decomposition and some discussion of them with the hope that 
it will give enough insight into the capability of the algorlthm and 
indicate to some extent the power of transformational analysis. 
•This work was carried out in the Transformations and Discourse Analysis 
Projectj University of Pennsylvania, sponsored by the National Science 
Foundation. 
1.0 Here, we will present a rather simplified description of an algo- 
rithm for transformational analysis (decomposition) of English sentences. 
Our purpose here is not to discuss the transformational theory 3 the full 
details of the theoretical formulations of the algorlthm 3 or ofthe 
grammar W. Rather, we will present a set of examples of the decomposition 
and some discussion of them with the hope that it will give enough in- 
sight into the capability of the algorithm and indicate to some extent 
the power of transformational analysis• 
I.i Transformations are certain relations among sets of sentences and 
in particular, it is possible to relate a given sentence to a set of 
elementary sentences (kernel sentences) by means of transformations. 
The kernel sentence forms (for English) are defined as the string of 
class marks N ~ ~ followed by one of the kernel Object strings: ~, N, 
~, ~P__~, __~, e~, D, A (m Noun; ~: ~ense/aux; ~: verb; ~: preposi- 
tion; D: adverb; ~: adjective; ~ : zero). Thus John bou~h't'a book; 
Nar~ will come etc. are kernel sentences. Each transformation is 
characterized by certain permutations, deletions or additions of spe- 
cific class marks or constants. In the resultant of a transformation 
one may look for s ubsequenceswhich remain~xg~n~ even when the re- 
sultant is subjected to further transformations. The 5asic features of 
the algorithm are 
a) stating the various invarlant sequences and 
b) formulating I) a grammar of such Invarlant sequences, 2) a corre~ 
spondlng recognition procedure, and 3) a systematic procedure for com- 
puting the kernel sentences as well as other kernel-like sentences and 
the corresponding transformational history. 
It should be emphasized tha t i t is not assumed and also not im- 
pl~ed in the algorithm that any kind of prior analysis (either strlng 
analysis or constituent analysis) is requlred as a prere~uisltelfor 
the present algorithm. 
~Such a detailed description will appear later elsewhere. 
++,+ ,..+ ++..., 
/i/+\~ 
1.2 Transformations are initially defined on kernel sentence forms. 
Howeverj they work on certain other sentence forms which are not kernel 
sentence forms. Thus a transformation is completely defined by first 
deflnlng it on a suitable kernel sentence form(s) and then extendlng 
• the ~omaln of the transformation to other "sentence forms. This exte~.- 
slon which contains infinitely many sentence forms can be represented 
by+ first listing a finite number of sentence forms in the extension and 
all the remaining sentence forms in the extension are obtained by cer- 
tain recurslve rules (see the i-llsts in 1.3). 
1.3 A unan~.z, transformation transforms one sentence form into another 
sentence form and a bi~nary transformation transforms a pair of sentence 
forms into another sentence form. Each unary transformation defined on 
a sentence form may be represented by a sequence of class marks consti- 
tuting another sentence form. Most binary transformations can be de- 
fined as interruptions of certain unary transformation sequences at 
stated positions by certain other sequences of class marks. These In- 
terruptlng sequences are not sentence forms but are deformatlons_-o~ sen- 
" -------"--'se ~ o~ <+ tence forms corresponding to the ~ sentence form of the binsry trans- 
formation. For example, John was detained by the old woman decom~ses 
into woman detained John and woman t be old with a passive transforma- 
tion on the first kernel and a binary transformation on the sentence 
John was detained by the woman and the kernel sentence woman t be old. ~ 
The sentence form corresponding to the passive transformation, 
N t be en V by N is then interrupted by the sequence AN before the last 
AN is a deformation of the kernel sentence form N t be A symbol. -- 
which is the second sentence form of the binary transformation. The 
resulting sentence form is thus N t be en V by A B,. In the resulting 
sentence form the shared symbol N appears only once. Such a symbol 
which two transformation sequences • share (or on which t_hey overlap) 
be ignore here the article th.._~e for simplicity. 
a. 2 -- 
wlll be called a residue of one.sequence with respect to another. 2 In 
addition to the transformation sequences which are sentence forms, and 
the interrupting sequences (deformed sentence forms) which correspond to 
most binary transformations, there is yet another type of interrupting 
sequences (again deformed sentence forms) which correspond tonominal- 
izatlons. For example consider: the book was written by Brown and 
John's travel to Italy was descrlbed by Mar~. In the second sentencej 
the kernel sentence John travelled to Italy is mappedonto the object 
of Mary described before the resultant undergoes the same passive which 
acted on Brown wrote the book giving the first sentence. N's nV P N 
(John's travel to Italy) is a nominalizati0n which appears in many dlf- 
ferent transformations and carries in them the associated kernel into 
one of the positions which could be occupied by a noun. For each 
transformation sequence in each intersymbol position we llst all in- 
terrupting sequences (including both the second and the third kind of 
sequences as dlscussedabove). Of course, the interrupting sequences 
have their own interrupting sequences, etc. These Intersymbol inter- 
rupting lists will be called i-lists. 
2. A sketch of the algorithm 
2.0 As stated in i, in order to define the set of all transforms we 
need a set Of sequences of class marks (or class mark-llke symbols) 
which has 3 subsets. 
1. Sequences each of which corresponds to a sentence form (e.g. the 
passive sequence N t be en V by N); 
2. Sequences each of which represents a deformed kernel-form and Is 
not a sentence form, but when inserted between specified neigh- 
boring symbols of a sequence of the first set~ preserves the 
character of the sentence form (e.g. _~j en V N); 
3. Sequences each of which represents a deformed kernel-form and is 
~he concept of the residue can be extended to shared sequences as well 
as sequences which replace a given symbol in another sequence. The 
term carrier is used in this context. This device has been extensive- 
ly used in this algorithm. 
-3- 
not a sentence form, but, when substituted for a symbol in a se- 
quence (of ~et I or 2 or 3), preserves the character of that se- 
quence (e.g. er V or/L , n A of N). 
There are also rules for inserting sequences from the second set 
into other sequences or into sequences of the third set, without 
changing the character of either. 
All insertion or replacement rules are stated in the interruption lists 
appearing between every pair of adjacent symbols of each sequence. 
Most of the sequences in the first set represent unary transforma- 
tions of kernel forms. Many are extended (often by permitting the 
replacement of certain symbols with selected sequences from the third 
set) to include analogous unary transforms of kernel-llke forms. 
The second set of sequencesj together with the rules of their in- 
sertion in the sequences for unary transformations, account for most 
of the binary transformations. Other binary transformations are rep- 
resented by replacement in pairs of class marks in unary transformation 
sequences by members of the third sets most of which consist of nomln- 
alizations. 
An arbitrarily long English sentence form can be seen as composed 
of a finite number of such sequences recurstvely embedded in one 
another. 
2.1 Corresponding to the above three subsets of sequences and their 
mutual embedding rules, we recognize three sets of strings. Each 
string is a program for comparing one of the sequences with a portion 
of the analyzed sentence form of the data. The program is equipped 
to permit interruption by other such programs according to the i-lists 
of the sequence. Each string, when entirely matched by a segment Of 
data, replaces that segment with the carrier of the string. The car- 
rier is sometimes null. In strings from the second set it is usually 
the residue of the binary insert (e.g. the center symbol of a noun 
phrase: N of AN, of~,etc.). In strings from the third set 
the carrier is a class-mark-llke symbol which, by replaclng a class- 
mark in a formderlved from a kernel form I extends it to one simi- 
--4 -- 
larly derived from a kernel-llke form. Let the carrier be~\[nV~\] for a 
noun phrase built around an nV. The extended passive form: 
N\[or~\[nV\]\] t be en V by N represents the form of the sentence John's 
travel to Italy was described by Haryas soon as the carrier of the 
str~ng replaces in the data the nomlr~ ! segment John's travel to Italy. 
The carrier from all strings in the first set is s, a symbol of a well- 
formed sentence. 
The program'of eachl string, whose sequence is a deformed (Or trans- 
forme~) kernel or kernel-like form, reconstructs that form for decom- 
position and attaches £o it a labe___~l descrlptive of the deformation (or 
transformation). The result of a decomposition is a set of kernel or 
kernel-llke sentences with labels. Some of the kernel sentences are / 
incomplete and have blanks in them because a transformation may de t ~ ....... 
elements. Some kernel-llke sentences may contain, instead of a wo/rd) 
a class-mark-llke symbol (e.g,~) with a reference.to aprevlous com- ) 
ponent of the decomposition. If that previous component is a kernel 
sentence (with or without blanks), then the label (describlr~ the de- 
formation) with the kernel-llke form (containing the reference) with 
its label, together constitute a description of the transformation un- 
dergone by the component kernel sentence. If the previous component 
itself is a kernel-llke sentence with a reference in turn to another 
component, both'kernel-llke sentences and all three labeis constitute 
the description of the transformatlonundergone by the c~,nponent ker- 
nel sentence ultimately referred to s etc. 
If the symbol x appears, instead of a word, in a kernel or ker' 
nel-llke sentence, it replaces a regular noun there. It is intro- 
duced in the sentence as a carrier from a nominallzation such as a 
teacher of Latln, .the driving instructor, etc. The same x must ap- 
pear in two or more sentences of the decomposition •(onewhere the 
nominal stands for a noun, and one in the sentence of whlch the nom- 
inallzatlon is a deformation, e.g. x - teach Latin). Which x's re- 
quire identical substitutions is discoverable, because each x has in 
a sharp bracket (< >) the names of every prevlous llne in which the 
same x appeared, often no actual substltutionls posslble and the x 
serves only to identlfy ~ with each other, two or more blanks in differ- 
ent components. The substitution of the noun replacing N for x in lines 
aj bj . . . d is implied when one kernel-llke component has the form 
I t be x ~/a~ b . . . d> . 
2.2 The three sets of strings (programs) constitute the major portion 
of granmmtlcal material in the algorithm. Another body o£ such material 
is the dlctlonar~. • 
The dictionary associates to each English word a symbol representing 
the wordts grannnatlcal class~ together with markers of certain addition- 
al characteristics the word may reveal by restricting its environment in 
the sentence. Some words may occur in more than one role and have there- 
fore several equlvalents in thedictlonary. (e.g. the word labor should 
be given four different class marks: present tense Vs V(untensed verb)j 
(nomlnallzatlon deslgnatlngthe activity of laboring)~ er'V (nominal- 
Izatlon designating the actor(s), possibly laborers in aggregate)). 
:al 
The dictionary for Transformation^Crannnar must carry far more de- 
tails than is needed for the String Analysis alone. Thus for example 
the transformational analysis must be able to discover in Johnts sleep 
not only a nouh phrase~ but also the incomplete kernel sentence John- 
sleep ~ which underlies each transformation containing such a noun 
phrase. Hence• the class marks: nV (sleep)j ~ (sharing) vA (brav- 
e y), (teacher), eeV (e loyee), (brotherho L aV (helpful) 
and several others. 
~:AV-entrY in the String Analysis dictionary contains information 
/about: the L~nd of objects required by the verb V. An nV may require 
objects differ/~t from its V and this must be indicated (e.g. th~ 
attacked the enemy vs. they made an attack " on the enemy). 
Noun phrases like n V, IngV, etc. can occur in place o£ a sen- 
tence object or a subject of a sentence but only when it is organized 
', \ . around a verb requiring such subjects or obJectsj and such, verbs are 
marked accordingly in the dictionary. 
The subject and object restrictions for a verb or a verb-related 
-6- 
wordare recorded in pairs~because they are not mutually independent. 
(0-" is the label for a subject (Z) requirement; _.~_~for an object (~) 
requirement of a tensed or untensed verb and some in_~ occurrences; 
t~nV labels an object requirement of nV-nominallzatlonj ~ingV those 
of an i~ V-nomlnalizatlonl etc. When neededj~ I is distinguished 
from ~ 2 (which usually is the same as the corresponding ~-~) to mark 
the form assumed by the object when it precedes the verb or verb re- 
lated word (compare for instance house construction with construction 
of house where c~ nVl (the same as o~ ) is N, while ~ nV2 (the same 
as nV ) IsP \[of\] N).) 
The analysis is preceded by a replacement of the Words in the 
sentence by corresponding entries in the dictionary. 
2.3 The process of analyzing a sentence begins in postulatlng (in 
turn) all those strings in the grammar which may occur at the beglnnlngP--q~o//~ 
of a sentence (and whose initial symbol is the s~me as the first symbol 
in the data). (See i I of #30). Each verified postulate forces other 
• postulates as its consequences I until the termlnal period of the sen- 
tence is found which is consistent with a hypothesis. It is qulte 
likely that an analysis will produce more than one correct reading of 
a sentencej because structural ambiguity is even more frequent in 
transformational grammar than it is in the mere s~ring analysis. 
-7- 
3. Examples of decomposition 
Four examples of decomposition obtained by the algorithm follow. 
These examples are intended to exhibit the power of the algorithm. 
It is posslble I without changing the algorithm I to increase the 
power and depth of the analysis by incorporating more details about 
transformations as they become available by adding either new trans- 
formation sequences or adding new classes and new co-occurencerestrlc- 
tlons in the dictionary or both. 
Among the various issues which are nowrecelvlng further atten- 
tlon, some are as follows: a) a better characterization of nF-nouns 
and the underlylng kernel sentences in terms of which the modifiers 
can he explained (e.g. school prlnclpa~ (example 3), French teacher; 
~ueen~ etc); b) the relation of classifier nouns to each other 
and their kernel positions with respect to'thelr modifiers (e.g. organ~ 
~e~¢er Ic chemistryj helpful trip, friendly ~e~j etc.); ¢) precise relatlon 
of constants (e.g. hi__ss~ both in example h) or classifier nouns with 
a definite artlcle to other nouns or phrases for which they are a re- 
placement. 
Examples: The first column lists the kernel sentences or kernel-llke 
sentences (or intermediate resultants). The second column glvesthe 
rest of the transformational history. Here the names as stated are 
partial in the sense that the corresponding strings do not always 
correspond to complete transformatlal sequences as discussed previously. 
ID 
Text: • The fact that John is a 
T N that N pres.be\[3\] T w 
his 1lie here unbearable. 
R'B nV D aV 
• stranger makes 
N present V\[3\] I 
.~I indicates here 3rd person. 
lo 
2. 
3. 
Decomposition: 
Kernel or Kernel- transformation 
llke sentences ~(partlal .@mes) carrier 
John pres.be stranger (a) container 2 noun: N w that S Nw. ~/l> 
He- live here; N~'-nomlnaliztion; g's n~ "1 "N <'2> 
- cannot bear~(2> adJectlvlzatlon : aV ~ ~3> 
N < l>pres, make'N< 2>AC3> container: N V NA S w w 
-2. 
Text: Our algebra teacher 
Rls N erV 
was requested by the school 
past be \[3\] enV by T N 
prlnclpal to interview 
nF Co V 
Decomposition: 
Kernel or Kernel- 
llke sentences 
a woman candidate from Swarthmore. 
T N N P N 
transformations 
(partial .names) carrier 
first reading: 
1. x - teach algebra 
2. We- have x~l> 
3. x - heads 3 school 
~. woman-'V P candidate (a) app app 
-a(V pp= be; Papp = ~) 
5. candldate-be from Swarthmore noun, right modified: •candidate "i P Na 
6. x <1,27 - interview . .-'~ passive of container: S 
candldate<~,5)~ N t VwN infinitive 
7- x (3~pas~ request x \[ <1,2  (6> -) 
x-nomlnallzation: GerV x ~i> 
• left modified noun: N's N x <1.,2> 
x-nominallzation: NnF x 43> ,., 
compound noun: NIN 2 candldate<~2 
k 
%oughly, container forms are sentence forms In which 1) there is a 
verb (Vw) requiring a sententlal subject or a sententlal object or 
both or 2) there is a noun (N w) or fidJectlve (Aw) requiring sen- 
tentlal complements. 
~eads is a V for nF principal as found in dictionary. -- appropriate 
-9 - : 
!Kernel or Kernel- 
llke sentences 
second reading: 
I. 
2. 
3. 
transformations 
.(partial names) 
x - teach us algebra 
x - head school 
woman - V P candidate (a) compound • noun: app app 
4. candidate-be from Swarthmore noun I right mod' 
ifled: NIPN 2 
~. x~J> - interview candidate ~passlve ofcon-  talner: 
 )Jlnfinitlve 6. x<2>past 
request x <1%2< , 
° • 
Text: 
x-nomlnallzation: flerv 
x-nominalization: NnF 5N2 
Accident insurance of an employee by 
N nV P T eeV P 
protects both. 
present, V: \[3\] Q 
Decomposition: 
Kernel or Kernel- 
~llke sentences 
first reading: 
I. - - employ x 
2. x - employ him 
3. x <2>- insure x < l>(an) 
P accident .app 
~. N ~3~ present protect both 
his employer 
R's erV 
second reading: 
1. - - employ x 
2. x -.employ - 
3. he - ~ve x /~2> 
h. x <2> - insure x <OCin) 
P accident app 
~. N ~h> present protect both 
Note: 
he 
Text: 
carrier 
x <l> 
x ~2> 
candidate 
<3> 
candidate z3, > 
S 
transformation 
(partlalnames) carrier 
x-nomlnallzation: eeV x ~I> 
x-nomlnalizati0n: erV ..x <2> 
~-nomlnallzatlon: nV N\[nV+~ 
• ÷z\] <3) 
container: N t V N W S W 
x-nomlnallzatlon: eeV 
x-nomlnalizatlon: erV 
left modified noun: 
• N'sN 2 
~-nominallzatlon: nV 
container: N t V N W 
. <l> 
x <2> 
X 
<2,3 > 
N\[nV+ ~ 
÷z\] <3> 
s 
The analysis would reach even deeper if the words ~hls and ~b°th 
were treated as reference words leading to a substltutlonj e.g. 
of x <I.__.~> for he, x ~I> and x <2> for both. 
Crop sharing between the tenant and the land owner 
N Ing V P T N and T N err 
Q may replace N. 
-10- 
Is 
present be \[3\] T 
organized labor. 
enV erV 
Decomposition: 
Kernel or Kernel- 
like sentences 
an economic arrangement unsatisfactory to 
aN nV aV P 
I. X - own land 
2. tenant (the) and x ~l>(the)- 
share crop 
3- - - arrange -; P economy app 
h. x - labor - x-nominallzatlon: erV 
~. - - organize x ~h> left modified noun: 
enVN 
6. ~<3> -not satlsfy x <~,5> q right modlfled:~aV 
7- N<2> present be'N- ~3,6~ (an) container: N'~t be'N 
transformations 
~(partlal names) carrier 
x-nomlnallzatlon: erV ..x ~I> 
"~-nomlnallzatlon: Ingv N\[ingVd- 
• ~~2> 
N"~-- nomlna llzat ion: nV N\[nv\] 
~3> 
x< ~,5> 
IS 
k. AnillustratiQn of the procedure 
Example 5 John is a good story teller 
This example illustrates the process of analysis in some detail. 
Because of space limitations for this paper a rather simple structure 
had to be chosen for this purpose. A short dictionary of the words in 
the sentence has been prepared and also a small set of grammar strings 
in provided for this illustration. Both were greatly simplified so 
that rich grammatical material will not obscure the demonstratl6n of 
the choice of hypotheses, their verification or rejection, the use of 
the carrler~ changes of levels in analysis and the exploration of al- 
ternative readings. 
The analysis always begins with the strlng ~30 postulated. A 
decomposition ends I when the program associated with thls~trlng is 
finished. All possible sentence beginnings are included in i I of ~30. 
--After- t-hee-nd of--~30-aiternatfve-decompositlons are sought. 
When a new string is postulated on the basis of an i-llst of 
.... another strlngj the verification of thenew string takes place in the 
next level of a push-down memory~ so that the state of computation of 
the suspended string is not affected. 
Whenever two or more alternative paths open up for the analy~s~i~sj 
each must be pursuedto a successful completion or until failure occurs. 
(The analysis must produce every possible decomposition of a structur- 
ally ambiguous sentence). In our analyslsj different paths are pursued ~ 
- 11 - 
serially. Every time an inspection of i-lists allows more than one 
hypothesls~ .one is chosen~ while a list of the remaining ones to- 
gether with all relevant positions of the.memory goes on top of 
another push-down storage. The contents of that •storage is examined . 
after the end of the chosen path. The analysis ends'after all pos- 
slble paths have been explored and thlsstorage is empty. ~n the 
example of analysis given herej we markbypassed open branches by 
asterisks on the left margin and their resumption by slmilar aster~ 
Isks encircled. ~ 
Dictionary used in Example ~. 
John - N \[proper, hu~anj singular\] . 
Is present be \[3~; ~r-: N\[or x\];¢~= N{A/PN/D. o-: S\[nV/inO\]; to. ~, ~, ~\[nV/ingv\] etc.\] 
a - T\[a\] 
good - A \[A-ly =well\] 
story - N 
teller- erV \[& : human, count; ~J=~erVl: N/N\[nV\]/~;~JerV2; 
PN\[or x or N: nV .~ 
Grammar Strings used in example 5- 
Nominal strings (each gives a noun-like carrier): 
I° T\[ the/a/an\] ll i N\[or x; or~: nV/ingV/nA/nl~\] 2 
- 
Dame= 
kernel: 
carrier: N\[the article\] (as matched) 
2. ~x i .\[orx\] 
i I 2,~,5, " 
i 2 - 
name: . left modified noun: AN 
kernel: N - be A 
carrier: N< address of kernel> 
~eslgnatesthlrd person, 
- ZZ~ 
. 
. 
e 
I0- 
Ii- 
N\[ object \] 
i I - 
.~ - 
• it. # x\]i2 A N\[nV/ingV; or 
i 1 .3, 
, i 2 - 
name: left modified nominal: A N \[or x\] 
addition to kernel of'N : ; A-ly (Alas matched) 
carrier: ~ ~as matched from data) 
i I err 
i I - 
name: x-nomlnallzatlon 
kernel: x-V- • 
carrier: x \[subclasses required from subject of V\]~address 
of kernel> 
i I i 2 erV 
name: x-nominalization 
kernel: x- V N '(N,V as matchedfromdata), 
carrier: x \[subclasses required from subject .of V\] ~address 
of. kernel> 
i 1 , i 2 nV 
• nv 
i 1 - 
• t 2 - 
name: N-nomlnallzatlon: nV\[+~ \] 
kerne I: f/n " 
carrier: ~n;\] v <a~dress of kernel~) 
Object Strings: 
.\[or x\] i: 
I I - • 
name- object 
contribution to kernel in carrier 
carrier: ~\[N\] 
il N\[or x\] P 
i I - 
i 2 1,2,~,9 
i s - 
(as matched from data) 
(N as matched in data) 
i 2 i 3 N\[or x\] 
w 13 - 
name - object 
contribution to kernel in carrier 
carrier:• ~\[N P N\] 
=. 
(N P N as matched in data) 
Sentence Strings 
20 - N t V g 
i I - 
i 2 - 
i 3 - 10,11,1,~,3,~,~,6 
% - 
21- 
DaJme: 
kernel: 
carrier: S < address of kernel> 
11 12 13 ~- i~ 
N t be N 
t 1 - 
i 2 - 
- 1,3,6 
name: ~containing "be" : N is N 
kernel: N t be N 
carrier: S 4 address of kernel> 
identity of kernel form: N t V 
N t V ~ (as found in data) 
(as matched from data) 
Monitor String 
i I i 2 i B 30 - " S " 
i I 1,2,3,~,~,6,20,21 
i 2 - 
i 3 - 
Zllustratlon of the process of analysis: 
Data:. N\[John\] pres.V\[3,be\] T\[a\] A\[good\] N\[story\] erV\[teller\]. 
~30:= .° S . (levell) 
S # N 
i I of 30 allows the following strilgs beginning with N to interupt 30 
here: ~,20. Try 20, mark ~ for the branch opening with 9 on level 2. 
W Data: N\[John\] pres V\[3,be\] Tie\] A\[good\] N\[story\] er V\[teller\] 
~20 : N tV ~ (level 2) 
N=N\[John\] 
t =present 
V=V b_e accepts Joh____nn as subject. For a human subject, the ob- 
Ject cannot be~ 6(in this simplified grammar). The verb b~e 
- 14 - 
rejects object form of ~Ii. 
R ~ T Among the remaining strings of i 3 
only 1 starts by T. 
Data: T\[a\] A\[good\] N\[story\] 
~i : T N 
T~T N~A 
f I of i has 2,3 beginning wlthA. 
bypassed 3. 
of 2o (1,2,3,a,%10) 
erV\[teller\]. (level 3) 
Try 2, mark ~.__~ for 
Data: A\[goodJ N\[story\] er V\[teller\]. (level ~). 
A---A 
~/ N=N note:i I of 2 has string 5 beginning withN. 
Mark ~-~ for the bypassed branch, 
end 2. kernel I: story-be good. Resume I. 
Data: ~\[story\] ~l> erV\[tener\]. (level 3) 
continue ~I 
$/~ N=N note:i I of I has string ~ beginning with N. Mark 
open branch ~-~. end 1. Resume 20. 
Data: N\[story\] < 1 > (a) er y\[teller\]. (level 2) 
contfnue~20 
of the strings from i 3 allowed by object requirement of the 
verb b ej ~ and I0 begin wfth N. Try 10j mark ___ ~ for by- 
passed ~. 
D ta:  \[story\] (a) err\[teller\]. (level 3) #IO: N 
N= N 
end i0. Resume 20. 
Data: ~ \[N\[Story\]< I> (a) J erV\[teUer\]. (level 2) 
contlnue~20. ~=~ \[N\[storyL<l> (a)\] 
End 20. Ke~del 2: John pres. be story < I> (a). Resume 30. 
D.ta= s crY\[teller> ( evel l) 
continue ~3o 
S =S 
-~er 
There is no string in f 2 of 30 which begins with er. Resume the near- 
est open branch: #~ at level 3- Erase mark ~ (kernel 2 is also 
erased) 
~ Data: N\[story\] ~I> (a) erV\[teller\]. (level 3) 
~5 : N erV 
N=N 
er • er 
V = V story ls a proper object f~r teller 
- 15 - 
End ~. Ker~e~ 2: x - tell story <I> (a). Resume 20. 
/ 
Data: x < 2> . (level 2) 
contlnue~20 
The only string beglnning with x among those of i 3 allowed as 
object of be is I0 
Data: x < 27. (level 3) 
~I0: N\[or x\] 
• N ~x 
end of I0. Resume ~20. 
Data: ~\[x <2>\] 
contlnue~0. 
= ~ \[x\[humanj ct., slngular\] < 2>\]. Howeverj the verb be with 
a count-noun subject requires from a noun object an ar- 
ticle or an artlcle-replacer. This lackingj the current 
branch fails, the branch marked ~ is reopened with #9 
onlevel ~. (Kernel 2 of the fai-~ng branch is erased.) 
Erase ~+~-~. 
Data: N\[story\] erV\[teller\]. (level h) 
~ : N erV 
N = N 
er = er 
V= V story is appropriate object for teller. 
End ~. Kernel 2: x - tell story < I~'. Resume I. 
Data: x < 2 >. (levei 3) 
continue ~i 
N=x 
end 1. Resume 20. 
Data: 
continue ~30 
S=S 
End 30 
Prin t output: 
Data: x 
continue ~x 
only one 
Data: 
~I0 : 
< 2>(a) . (level 2) 
20 
string beginning by x can interrupt here;it is I0. 
CLevel 3) ~\[or x\] 
N~ X 
end IO. Resume 20 
Data: fi\[x < 2> (a)\] . (level 2) 
contlnue~20 fl = fi 
end 20. Kernel 3: John pres. be x ~ 2> (a). 
S .--=-- (level I) 
Resume 30. 
1. story - be good (left modified noun) 
2. x - tell story < 1 > \ (x-nomfnalfzatfon: flerV) 
3. John present be x < 2> (a) (identity of extended NtVN) 
"- 16 - 
Are there any branches open? 
Erase ~ek. 
Yes, ~-~ at level ~. #~ will be tried. 
- Data: N\[story\] erV\[teller\]. (level 5) 
• ~5 : N erV 
N=N 
er = er 
V = V ~ is appropriate object of teller 
end ~. .Kernel la: x - tell story. Resume 2. 
Data: xXl>. (level~) 
continue -~ 
N-x 
End 2. Kernel 2a: x ~I~- be good. Resume I. 
Data: 
continue ~i # 
N----x 
end I. Resume 20. 
(level 3) 
Data: x (1,2> (a). (l.~el 2) 
continue 
The on!y string allowed to interrupt here is 10. 
Data: x ~1,2~ (a). (level 3) ~Io: N\[orx\] 
N=x 
End 10. Resume 20. 
End 20. Kernel 3a: John pres. be X ~It2~ 
Data: s ~3>. (level l) 
continue 30 
S~S 
e-- • 
end 30. 
(a). Resume 30. 
print output: I. x - tell story (x-nomfnallzation: ~erV) 
2. x ~i> - be good (left modifier noun) 
3. John pres. be x ~1,2> (a) (identity of extended 
Are there any branches open? Yes, ~l~ at level ~. 
(To abbreviate, we will,just say t~-t--this branch will be very much llke 
the last one, except that, due to the difference between strings 2 and 
3~ it will give the output: 
- 17 - 
1. x- tell story; well (x-nominalication: ~erV; 
left modified nominal) 
2. John pres. be x < I> (a) (identity of extended 
NtVN) 
The last open branch, marked ~ fails immediately.) 
. 
2. 

References 

3oshij A. K., Hi~1D.j "String representation of transformations 
and a decomposition procedure"~ Part I and Part IIj Transforma- 
tions and Discourse Analysis Project PaperIUnlverslty of 
Pennsylvania; Dec. 1965. 

3oshi s A. K.~ "Transformational analysis by computer with some:i 
appllcation~'j Presented at the National Institute of Health 
Seminar on Computational Linguistics, Bethesda, Oct. 1966~: 
(To be published). 
