I. Introduction 
To a great extent, the lack of large-scale effective solutions to 
problems of computatlonal linguistics can be traced to the lack of an 
adequate linguistic theory that could be used as a framework for computa- 
tional work. Most of the linguistic theorizing that has taken place in 
the United States has been done under the banner of Transformational 
grammar. Fundamental to transformational theory is the sharp distinc- 
tion between 'competence' and !performance'. 
This distinction between competence and performance provides for 
transformatlonallsts the platform from which to make their statements 
about transformations. 'Competence', according to Chomsky is what the 
speaker-hearer knows about his language, as opposed to his use of that 
knowledge, labeled 'performance'. Chomsky includes in his discussion of 
what a 'performance' model should do, factors such as memory limitations, 
inattentlon~ distraction, and non-linguistlc knowledge. He thus leaves 
for 'competence' the formalization of linguistic processes representative 
of the speaker-hearer's knowledge of the language. 
This relegation of competence makes a basic mistake however. It 
is necessary to differentiate the problem of formalization of linguistic 
knowledge and processes, i.e., competence, from the simulation of lin- 
guistic knowledge and processes, which we shall call 'simulative perfor- 
mance'. There is a difference between the simulation of knowledge and 
processes ('simulative performance') and the simulation of actual ver- 
J 
bal behavior (Chomsky's 'performance'). It is here that we must speak, 
as chomsky does, of the ideal speaker-hearer. Clearly the ideal speaker- 
hearer is not inattentive or distracted. He does however have memory 
-I- 
limitations and non-linguistle knowledge. This is certainly what must 
be simulated as an inclusive part of linguistic theory. The kind of 
theory of 'performance' of which Chomsky speaks may well be in the far 
distant future to which Chomsky relegates it (1965). However, a theory 
of simulative performance is not so far off. It would seem very reason- 
able that the possibility of the construction of a linguistic theory that 
both accounts for the data and does this in such a way as to appear to 
be consonant with the human method for doing so, is not so remote. Clear- 
ly, such a theory must deal with non-linguistic knowledge and problems 
of human memory as well as the problems that Chomsky designates as 'com- 
petence'. Thus, it seems that the sharp distinction between competence 
and performance is artificial at best. In particular, after elimina- 
tion of some of the behavioristic problems such as distraction, we can 
expect to find a linguistic theory that is neither one of 'competence' 
nor 'performance' but something in between and therefore inclusive of 
both. 
Chomsky (1965:139) has stated: 
'thus it seems absurd to suppose that the speaker first 
forms a generalized Phrase-marker by base rules and then 
tests it for well-formedness by applying transformational 
rules to see if it gives, finally, a well-formed sentence. 
But this absurdity is simply a corollary to the deeper ab- 
surdity of regarding the system of generative rules as a 
point-by-point model for the actual construction of a sen- 
tence by a speaker." 
We could, on the other hand, attempt to formulate a system of 
rules that are a point-by-point model for the actual construction of a 
sentence by a speaker. Furthermore, we might expect that that system 
could also be a point-by-polnt model for the actual analysis of a sen- 
tenee by a hearer. These claims, however, would be largely unverifiable 
-2- 
except by the use of computers as simulative devices. 
Chomsky (1965:141) has further stated that: 
'She gralmnar does not, in itself, provide any sensible 
procedure for finding the deep structure of a given sentence, 
or for producing a given sentence, just as it provides no 
sensible procedure for finding a paraphrase to a given sen- 
tence. It merely defines these tasks in a precise way. A 
performance model must certainly incorporate a grammar; it 
is not to be confused with grammar." 
Insofar as the notion of a performance model here can be taken as 
being somewhere between Chomsky's notion of competence and performance, 
our notion of grammar also lies somewhere between Chomsky's notion of 
a grammar and the incorporation of a grammar. 
-3- 
II. Conceptual Dependency 
The Conceptual Dependency framework (see Schank \[ 1969\] ) is a 
stratified linguistic system that attempts to provide a computational 
theory of simulative performance. The highest level of the stratifi- 
catlonal system (similar to Lanab \[ 1966\], Sgall \[1965\] and others) em- 
ployed by the Conceptual Dependency framework is an interlingua consis- 
ting of a network of language-free dependent concepts, where a concept 
may be considered to be an unambiguous word-sense, (except see Schank,\[1968\]). 
(The notion of dependency used here is related to those of Hays (1964) 
and Klein (1965), however, the dependencies are not at all restricted 
to any syntactic criterion.) The graumaar of a language is defined by 
the framework as consisting of Realization Rules that map conceptual 
constructs into syntactically correct language on the 'sentential level' 
The linguistic process can be thought of, in Conceptual Dependency 
terms, as a mapping into and out of some mental representation. This 
mental representation consists of concepts related to each other by var- 
ious meaning-contingent dependency links. Each concept in the inter- 
lingual network may be associated with some word that is its realizate 
on a sentential level. 
The conceptual dependency representation is a linked network that 
can be said to characterize the conceptualization inherent in a piece 
of wrltten language. The rule of thumb in representing concepts as 
dependent on other concepts is to see if the dependen t concept will fur- 
ther explain its governor and if the dependent concept cannot make sense 
without its governor. 
For example, in the sentence, '~he big man steals the red book 
-4- 
from the girl." the analysis is as follows: 'The' is stored for use in 
connecting sentences in paragraphs, i.e., 'the' specifies that 'man' may 
have been referred to previously. 'Big' refers to the concept 'big' 
which cannot stand alone conceptually. The concept 'man' can stand 
alone and is modified, conceptually by 'big', so it is realized in the 
network as a governor with its dependent. 'Steals' denotes an action 
that is dependent on the concept that is doing the acting. A conceptual- 
ization (a proposition about a conceptual actor) cannot be complete 
without a concept acting (or an attribute statement), so a two-way depen- 
dency link may be said to exist between 'man' and 'steal' That is, they 
are dependent on each other and govern each other. Every conceptualiza- 
tion must have a two-way dependency llnk. 'Book' governs 'red' attribu- 
tively and the whole entity is placed as objectively dependent on 'steals'. 
The construction 'from the girl' is realized as being dependent on the 
action through the conceptual object. This is a different type of de- 
pendency (denoted by 4). There are different forms of this 'prepositional 
dependency', each of which is noted by writing the preposition over the 
llnk to indicate the kind of prepositional relationship. (Although a 
language may use inflections or nothing at all instead of prepositions 
to indicate prepositional dependency, we are discussing a language-free 
system here and it is only the relation of the parts conceptually that 
is under consideration.) 
The conceptual network representation of this sentence is then 
as follows: 
from 
man ~ steals ~ book ~ , girl 
t t 
big red 
-5- 
The conceptual level works with a system of rules (shown in the 
Appendix) that operate on conceptual categories. These rules generate 
all the permissible dependencies in a conceptualization. Multiple 
combinatlon of conceptualizations in various relationships are intended 
to account for the totality of human language activity at the conceptual 
level. 
The conceptual categories are divided into governing and assisting 
groups: 
Governin~ Categories 
PP 
ACT 
LOC 
T 
Assisting Categories 
PA 
AA 
An actor or object; corresponds syntactically 
(in English) to concrete nominal nouns or noun 
forms. 
An action; corresponds syntactically (in English) 
to verbs, verbal nouns, and most abstract nouns. 
A location of a conceptualization. 
A time of conceptualization; often has variant 
forms consisting of parts of a conceptualization. 
Attribute of a PP; corresponds (in English) to 
adjectives and some abstract nouns. 
Attribute of an ACT; corresponds (in English) to 
adverbs and indirectly objective abstract nouns. 
Thus, the categories assigned in the above network correspond closely to 
their syntactic correlates: 
PP ~ ACT ~ PP ~ PP 
PA PA 
However, in the sentence, 'Visiting relatives can be a nuisance', the 
syntactic categories often do not correspond with the conceptual actors 
-6- 
and actions. The ambiguous interpretations of this sentence are: 
one PP 
(I) ~ ~ bother ~ one ~ ~ ACT ~ PP 
visit ACT 
relatives PP 
(Here we use the conditional present \[denoted 
by c\] form of the two-way dependency link, 
one of eight posslble tense-mood forms.) 
relatives ~ bother ~ one PP ~ ACT ~ PP (2) % $ 
visit ACT 
relatives PP 
(3) bother, one $ 
visit ACT 
one PP 
A conceptualization is written in a conceptual dependency analysis 
on a straight line. Dependents written perpendicular to the line are 
attributes of their governor except when they are part of another con- 
ceptualization line. Whole conceptualizations can relate to other con- 
ceptualizations as actors (\[i\] and \[3\]) or attributes (\[2\] where ~ in- 
dicates that the PP at its head is the actor in a main and subordinate 
conceptualization \[ ~ is the subordinate, written below the line\]). 
The Conceptual Dependency framework, at the conceptual level, is 
thus responsible for representing the meaning of a piece of written 
language in language-free terms. The representation is in terms of 
actor-action-object conceptualizations in a topic-cogent form. Thus, 
words that have many syntactic forms will have only one conceptual form. 
This is true interlinguistically as well as intralinguistically. The 
-7- 
meaning of a construction is always the consideration used in represen- 
tation. For example, 'of' in 'a chp of water' is realized as '~-~ con- 
talns X' where X is water. 
cup 
contains 
water 
Similarly, in 'John's love is good', 'love' is realized conceptually as 
X = loves ~Y. 
John 
~ good 
love 
t 
one 
In order to make this framework serve as a generative theory, 
semantics and realization rules must be added. The realization rules are 
used in conjunction with a dictionary of realizates. These rules map 
pieces of the network in accord with the granmaar. Thus, a simple rule 
in English might be: 
PP 
= Adj + N 
PA 
In facts the rules are not this simple since criteria of usualness and 
context enter into each application of a rule. These problems are dis- 
cussed elsewhere (Sehank \[1969\] ) and are not the point of this paper. 
The semantics that Conceptual Dependency employs is a conceptual 
semantics in that it serves only to limit the range of conceptualizations 
in such a way as to make them consonant with experience. The form and 
major content of this semantics is thus universal, but Since we are deal- 
ing with experience we are required to speak of someone's experience. 
-8- 
We will thus begin to talk about some arbitrary human's experience, or 
since we are dealing with a computer, we can talk of the systems' ex- 
perience. Thus, the conceptual semantics consists of lists of potential 
dependents for any given concept. These lists are listed with respect 
to semantic categories if there is a generalization that can be made on 
that basis. 
-9- 
III. The Parser 
The Conceptual Dependency framework is used for a natural language 
parser by reversing the realization rules and using the semantics as a 
check with reality. The system for analyzing a sentence into its con- 
ceptual representation operates on pieces of a sentence looking up the 
potential conceptual realizates. 
All conceptualizations are checked against a list of experiences 
to see if that particular part of the construction has occurred before. 
If the construction has not occurred, or has occurred only in some 
peculiar context, this is noted. Thus, in the construction 'ideas 
sleep', it is discovered that this connection has never been made before, 
and is therefore meaningless to the system. If the user says that this 
construction is all right, it is added to the memory; otherwise the con- 
struction is looked up in a metaphor list or aborted. The system thus 
employs a record of what it has heard before in order to analyze what 
it is presently hearing. 
In order for the system to choose between two analyses of a sen- 
tence both of which are feasible with respect to the conceptual rules 
(see Appendix) the conceptual semantics is incorporated. The conceptual 
semantics limits the possible conceptual dependencies to statements con- 
sonant with the system's knowledge of the real world. The definition of 
each concept is composed of records organized by dependency type and by 
the conceptual category of the dependent. For each type of dependency, 
semantic categories (such as animate object, human institution, animal 
motion) are delimited with respect to the conceptual category of a 
given concept, and defining characteristics are inserted when they are 
- I0- 
known. For example, concepts in the semantic category 'physical object' 
all have the characteristic 'shape' Sometimes this information is in- 
trinsic to the particular concept involved, for example, 'balls are 
round' 
The semantic categories are organized into hierarchical structures 
in which limitations on any category are assumed to apply as well to all 
categories subordinate to it. The system of semantic categories and a 
method of constructing semantic files is discussed more fully in Schank 
(1969). 
In the present system, the files are constructed by incorporating 
information derived from rules presented as English sentences. The 
program parses each of these sentences and observes which dependencies 
are new and then adds them to the files. 
As an example of the use of the conceptual semantics, consider 
the parse of 'the tall boy went to the park with the girl'. At the 
point in the parse where the network is 
boy ~ go ~ park 
t 
tall 
we are faced with the problem of where to attach the construct ~ ~tb girl. 
A problem exists since at least two realization rules may apply: 'ArT 
PR~P ~: 1 ~3; '~P P~EP ~P: ~2 ' The problem is resolved by the 
3 
conceptual semantics. The semantics for 'go' contains a llst of concep- 
tual prepositions. Under 'with' is listed 'anyl movable physical object' 
and since a girl is a physical object the dependency is allowed. The 
semantics for 'park' are also checked. Under 'with' for 'park' are 
listed the various items that parks are known to contain, e.g., statues, 
-ll- 
junglegyms, etc. 'Girl' is not found so the network (I) is allowed 
while (e) is aborted. 
(I) boy g go ~to park <with girl 
tall 
(2) boy g go <t=o park 
t 
tall ~with 
girl 
Although ,~'th girl' is dependent on 'go' it is dependent through 
'park'. That is, these are not isolated dependencies since we would 
want to be able to answer the question 'Did the girl go to the park?' 
affirmatively. In (2) the below-the-line notation indicates that it is 
the 'park with a girl' as opposed to another 'park'. Now it may well be 
the case that this is what was intended. The conceptual semantics func- 
tions as an experience file in that it limits conceptualization to ones 
consonant with the system's past experience. Since it has never encoun- 
tered 'parks with girls' it will assume that this is not the meaning 
intended. It is possible, as it is in an ordinary conversation, for the 
user to correct the system if an error was made. That is, if (2) were 
the intended network it might become apparent to the user that the 
system had misunderstood and a correction could easily be made. The 
system would then learn the new permissible construct and would add it 
to its semantics. The system can always learn from the user (as des- 
cribed in Schank \[1968\] ) and in fact the semantics were originally input 
in this way, by noticing occurrences in sample sentences. 
Thus, the system purports to be analyzing a sentence in a way 
-12- 
analogous to the human method. It handles input one word at a time as 
it is encountered checks potential linkings with its ~n knowledge of the 
world and past experience, and places its output into a language-free 
formulation that can be operated on, realized in a paraphrase, or trans- 
lated. 
Thus the Coneeputal Dependency parser is a conceptual analyzer 
rather than a syntactic parser. It is primarily concerned with expli- 
cating the underlying meaning and conceptual relationships present in 
a piece of discourse in any natural language. The parser described here 
bears some similarity to certain deep structure parsers (Kay \[1967\] , 
Thorne et al \[ 1968\] and Walker \[1966\] ) only insofar as all these parsers 
are concerned to some extent with the meaning of the piece of discourse 
being operated upon. However, the parser is not limited by the problems 
inherent in transformational grammar (such as the difficulty in revers- 
ing transformational rules and the notion that semantics is something 
that 'operates' on syntactic output). Also, the parser does not have 
as a goal the testing of a previously formulated grammar \[as does 
Walker (1966) for example) so that the theory underlying the parser has 
been able to be changed as was warranted'by obstacles that we encountered. 
The parser's output is a language-free network consisting of unambiguous 
concepts and their relations to other concepts. Pieces of discourse 
with identical meanings, whether in the same or different languages, 
parse into the same conceptual network. 
The parser is being used to understand natural language state- 
ments in Colby's (1967) on-line dialogue program for psychiatric inter- 
viewing, but is not restricted to this context. In interviewing 
-13- 
programs llke Colby's, as well as in question-answering programs, a 
discourse-generating algorithm must be incorporated to reverse the 
function of the parser. The conceptual parser is based on a linguistic 
theory that uses the same rules for both parsing and generating, thus 
facilitating man-machine dialogues. 
In an interviewing program, the input may contain words that the 
program has never encountered, or which it has encountered only in 
different environments. The input may deal with a conceptual structure 
that is outside the range of experience of the program, or even use a 
syntactic combination that is unknown. The program is designed to 
learn new words and word-senses, new semantic possibilities, and new 
rules of syntax both by encountering new examples during the dialogue 
and by receiving explicit instruction. 
-14- 
IV. Implementation 
The parser is presently operating in a limited form. It is 
coded in MLISP for the iPDP-iO and can be adapted to other LISP processors 
with minor revisions. 
Rather than attaching new dependencies to a growing network during 
the parse, the program determines all the dependencies present in the 
network and then assembles the entire network at the end. Thus, the 
sentence 'The big boy gives apples to the pig.' is parsed into: 
i) ~ boy 
t 
big 
2) boy ~ give 
3) gives ~ apples 
~) give <~ pig 
and then these are assembled into: 
boy ~ give ~ apples ~ pig 
t 
big 
The input sentence is processed word-by-word. After "hearing" 
each word, the program attempts to determine as much as it c~n about the 
sentence before "listening" for more. To this end, dependencies are 
discovered as each word is processed. Furthermore, the program antici- 
pates what kinds of concepts and structures may be expected later in the 
sentence. If what it hears does not conform wlth its anticipation, it 
may be"confused", "surprised", or even "amused". 
In case of semantic or syntactic ambiguity, the program should 
determine which of several possible interpretations was intended by the 
"speaker". It first selects one interpretation by means of miscellaneous 
-15- 
heuristics and stacks the rests. In case later tests and further input 
refute or cast doubt upon the initial guess, that guess is discarded or 
shelved, and a different interpretation is removed from the stack to be 
processed. To process an interpretation, it may be necessary to back up 
the scan to an earlier point in the sentence and rescan several words. 
To avoid repetitious work during rescans, any information learned about 
the words of the sentence is kept in core memory. 
The parse involves five steps: the dictionary lookup, the appli- 
cation of realization rules, the elimination of idioms, the rewriting 
of abstracts~ and the check against the conceptual semantics. 
The dictionary of words is kept mostly on the disk, but the most 
frequently encountered words remain in core memory to minimize processing 
time. Under each word are listed all its senses. "Senses" are defined 
pragmatically as interpretations of the word that can lead to different 
network structures or that denote different concepts. For example, 
some of the senses of "fly" are: 
fly I - (intransitive ACT): what a passenger does in an airplane. 
fly 2 - (intransitive ACT): what an airplane or bird does in the 
air. 
fly 3 - (PP): an insect 
fly 4 - (transitive ACT): what a pilot does by operating an airplane. 
fly 5 - (intransitive AcT--metaphoric): to go fast. 
fly 6 - (PP): a flap as on trousers. 
If ther~ are several senses from which to choose, the program 
sees whether it was anticipating a concept or connective from some spe- 
cific category; if so it restricts its first guesses to senses in that 
category. Recent contextual usage of some sense also can serve to prefer 
-16- 
one interpretation over others. To choose among several senses with 
otherwise equal likelihoods, the sense with lowest subscript is chosen 
first. Thus, by ordering senses in the dictionary according to their 
empirical frequency of occurrence, the system can try to improve its 
guessing ability. 
The realization rules that apply to each word sense are referenced 
in the dictionary under each sense. Most of the rules fall into cate- 
gories that cover large conceptual classes and are referenced by many 
concepts. Such categories are PP, PA, AA, PPloc, PPt, LOC, T, simply 
transitive ACT, intransitive ACT, ACT that can take an entire concep- 
tualization as direct object ("state ACT") and ACT that can take an 
indirect object without a preposition ("transport ACT"). In contrast 
to most concepts, each connective (e.g., an auxiliary, preposition, or 
determiner) tends to have its own rules or to share its rules with a 
few other words. 
A realization rule consists of two parts: a recognizer and a 
dependency chart. The recognizer determines whether the rule applies 
and the dependency chart shows the dependencies that exist when it does. 
In the recognizer are specified the ordering, categories, and Inflection 
of the concepts and connectives that normally would appear in a sentence 
if the rule applied. If certain concepts or connectives are omissible 
in the input, the rule can specify what to assume when they are missing. 
Agreement of inflected words can be specified in an absolute (e.g., 
"plural") or a relative manner (e.g., "same tease"). Rules for a 
language like English have a preponderance of word order specifications 
while rules for a more highly inflected language would have a preponderance 
-0- 
of inflection specifications. 
Realization rules are used both to fit concepts into the network 
as they are encountered and to anticipate further concepts and their 
potential realizates in the network. When a rule is selected for the 
current word sense, it is compared with the rules of preceding word 
senses to find one that "fits". For example, if "very hot" is heard, 
one realization rule for "very" is: 
vdry 
where the tags "0" and "i" indicate the relative order of the word sense 
in the recognizer and identify them for reference by the dependency 
chart; '~" means the current word. One rule for "hot" is: 
0 AA PA : 
-i0-i 
The program notices that "very" fits in the "-i" slot of the "hot" rule 
and verifies that "hot" fits in the "i" slot of the "very" rule. There- 
fore, the dependency suggested by the chart can he postulated for the 
network: 
hot (PA) 
very (AA) 
After the rules for two adjacent word senses are processed, other 
rules are tried, and more distant word senses are checked. 
Whenever a dependency is postulated, it is looked up in an idiom 
file to see if it is an idiom or a proper name and should be restructured. 
Thus, the construct: 
make 
up 
-18- 
is reduced to the single concept 
make-up 
This idiom will be detected by the parser even if several words inter- 
vene between '~make" and "up" in the sentence. 
After eliminating idioms from the network, there still may be 
constructs that do not reflect language-free conceptions. The most 
conspicuous eases are caused in English by abstract nouns. Most such 
nouns do not correspond to PP's but rather are abbreviations for con- 
ceptualizations in which the concept represented is actually an ACT or 
a PA. 
The program treats an abstract noun as a PP temporarily in order 
to obtain its dependents, because abstract nouns have the syntax not of 
ACT's but of PP's. After obtaining its dependents, the PP is rewritten 
as an entire conceptualization according to rules obtained from an ab- 
stract file. These rules also specify what to do with the dependents of 
the PP; they may be dependent on the entire conceptualization , dependent 
on the ACT only, or appear elsewhere in the conceptualization. 
By way of example, the sentence: 
Tom's love for Sue is beautiful. 
leads to the following dependencies; 
love (PP) love (PP) 
for 
Tom (PP) Sue 
After hearing "is", the program expects no mor~ dependents for "love" 
(by a heuristic in the program), so it checks the abstract file and 
finds rules for "love" including: 
-19- 
off,for : 
PP PP 
(a) (b) 
where "(a)" and "(b)" identify concepts without reference to senten~al 
order. The network Is now rewritten: 
Tom 
love 
t 
Sue 
where the horizontal main link represents "is", waiting for a right-hand 
concept. When 'beautiful" Is heard, the network is complete, giving: 
Tom 
~ beautiful 
love 
t 
Sue 
The network above may be realized alternative~ as either of the 
paraphrases: 
That Tom loves Sue is beautiful. 
For Tom to love Sue is beautiful. 
In conceptual dependency theory, connectives like "that", "for", "to"~ 
and "of" are cues to the structure of the network and need not appear 
in the network at all. The network above demonstrates such a situation. 
Conversely, portions of the network may be absent from the sentence. 
For example, the sentence: 
It is good to hit the ball near the fence. 
-20- 
is parsed as: 
one 
~ good 
hit 
t 
ball 
to 
fence n~arPlaee 
Here, "one" and "place" are not realized. Notice that the relevant 
realization rule for "it" is: - \] (al)Ione) 
If°r PP it be PA ACT: 2 ~a0) (al) 0 i 2 3 
The square brackets indicate optional words. The tags "(a0)" and "(al)" 
indicate that "for" precedes the "PP" but the whole phrase may occur in 
any position of the construct. "(al)!one" in the dependency chart means 
that if "(al)", i.e., "for PP", is omitted, and the subject of the action 
is not obvious from context, then the concept "one" is to be assumed. 
The conceptual network should reflect the Beliefs inherent in 
the original discourse in a language-free representation. The inter- 
linguistic file of conceptual semantics is checked to verify that the 
dependencies are indeed acceptable. This check is made after abstracts 
have been rewritten. 
After the five parsing steps are completed, the program proceeds 
to the next word. At the end of the sentence, it outputs the final net- 
work in two dimensions on a printed page or on a display console. 
-21- 
V. Examples of Algorithm 
Only a few of the relevant realization rules will be shown in 
the examples. 
Example I 
'John saw birds flying to California. 
Realization Rule Patterns 
Words (for posslble senses) 
John 1: (all PP patterns) 
saw I: (all PP patterns 
2: (tO see, past tense) 
PP ACT PP: "i ~ 0 ~- i -i 0 I 
PP ACT ~ PP: 2 ~ 0 ~ I -i0 Y 2 
Dependencies 
(rectangle around new dependencies 
~ohn ~"see\[~ Pe 
(note: "to"means "tense of ACT 
Number O) 
3: (to saw):.., etc. 
birds 1: (all PP patterns)  ying l:a) A -Ing: 1 o 
l:b) ACT PP ACT-Ing : 
0 i 2 
\[see ,- birds_~ 
~birds ~ ~l~ I 
But now there are two main links 
on one llne so go back and try as 
object of 'see'. 
I 
birds 
see ~ f~ly 
i: ACT to PPLoc-I ~ i 
-I 0 I 
2: to ACT ~ i 0 1 
etc. 
fly ~ PPLoc 
-22- 
California I: (all PPLoc patterns) 
Final output: 
fly <to California 
birds g 
John see ,- $ 
fly ton 
California 
Example 2 
'John saw Texas flying to California.' 
Words Patterns 
John I: (all PP patterns ) 
saw (as above) 
Texas I: (all PPLoc patterns) 
flying (as above) 
(rest as above) 
Final Output: 
John ~ see ~ Texas 
P John ~ fly<t=°California 
Dependencies 
J~nl ~ see!~ PP 
Texas ~ fly: rejected by 
semantics. Laugh and go 
back and try (Ib): 
J°hnFIsee 
"" p l John! \[fl d 
Example 
'Jane ate the hamburgers in the park.' 
Words , Patterns 
Jane I: (all PP patterns) 
ate i: (eat - past tense) 
PP ACT PP: -it~O0 ~ I 
-I 0 1 
Dependencies 
John ~ eat 
eat ~ PP 
e tc. 
-25 - 
Words 
hamburgers 
in the park 
Patterns 
i: (all PP patterns) 
I: ACT In PP: -i 0¢_ i 
-I O i 
2: PP in PP: -I 
-1 01 0 ~ 1 
3: PP ~2A~ in PP • -3 " " 0 1 LOC" 
-2 
-3 ~ ° -i 
i 
Dependencies 
eat ~ hamburgers 
eat ~ park: rejected by 
semantics 
hamburgers 
~in 
park 
set aside as unlikely 
(would accept if not other 
alternatives) 
P 
John ~ eat 
in II 
park 
-24 - 
VI. Examples of Parses 
'Flying planes can be dangerous.' 
planes g dangerous 
fly 
one 
g dangerous 
fly 
t 
plane 
'The shooting of the hunters was terrible.' 
hunters 
~ terrible 
shoot 
one 
~ terrible 
shoot 
t 
hunters 
'John, who was in the park yesterday, wanted to hit Fred in the 
mouth today'. 
today 
John ~ want 
John ~ hit ~ Fred park ~ be <~ mouth t ay of 
yesterday Fred 
'John was persuaded by the doctor in New York to be easy to please.' 
doctor ~ persuaded ~ John 
n one ~ please ~ John 
New York easy 
'The girl i llke left.' 
glrl ~ leave 
I ~ like 
"25 - 
Vll. Conclusion 
Before computers can understand natural language they must be able 
to make a decision as to preclsely what has been said. The conceptual 
parser described here is intended to take a natural language input and 
place the concepts derivable from that input into a network that expli- 
cates the relations between those concepts. The conceptual network that 
is then formed is not intended to point out the syntactic relations 
present and there is some question as to why any system would want this 
information. Although Chomsky's deep structures convey a good deal 
more information than just syntactic relations, it is clear that a parser 
that uses deep structures for output would be oriented syntactically. 
We see no point in limiting our system by trying to test out a previously 
formulated grammar. The output of a transformational parser, while 
making explicit some important aspects of the meaning of the sentence, 
does not make explicit all the conceptual relationships that are to be 
found, does not limit its parses with a check with reality, and most 
importantly is syntax based. The parser presented here is semantics 
based. We aver that the system that humans employ is also semantics 
based. It seems clear to us that our parser satisfies the requirements 
that a parser must satisfy and in so doing points out the advantages of 
regarding language from a Conceptual Dependency point of view. 
-26 - 
APPENDIX 
I. Conceptual Rules (permissible dependencies): 
PP * ACT; PP = PP; PP * PA; ACT ~ PP; ACT'~PP; 
PP PP ACT PA AA PA T IDC ACT 
~; t; t; t; ~; ~; ~; ,; ~;~. 
PP PA AA PA PA PP c~ ~ o 
II. Realization Rules 
There are about I00 of these rules, presently. 
here. 
A few are shown 
PP ACT : I ~ 2 ; PP ACT to ACT : 1 @ 2 ; PA PP : 2 ; 
i 2 1 2 3 t 1 2 ¢ 
i~3 I 
ACT PP PP : i "~2 ~-3 ; PP Prep PP : i 
12 3 1 2 3 ,'~'2 ; 
3 
PP ACT Prep PP : i ~ 2 ~4 for I ~2 ; PP PP ACT ACT: i ~4; 
l 2 3 ~ ~3 i e 3 4 2~3 
l 
ACT PP ACT-lug : ~ ; PP who ACT ACT : i * 3 
i 2 3 2~3 i ~ 3 % 2 
III. A Sample of the Conceptual Semantics for 'ball'. 
ball, 
inanimate motion object 
-27" 
~-PA PP 
size any in phys obj 
shape round on phys obj 
color any for phys obj 
texture usually smooth by place 
elasticity bounces of animal 
at no 
to no 
ACT 
specific 
motion object 
concrete 
any 
bounce 
roll, come, spin 
fall, hit ... 
begin, cause ... 
-28- 

References

i. Chomsky, N., Aspects of the Theory of Syntax, MIT Press, Cambridge, 
1~5. 

2. Colby, K., and Enea, H., "Heuristic Methods for Computer Understan- 
ding of Natural Language in Context-Restricted On-Line Dia- 
logues," Mathematical Biosciences, 1967. 

3. Hays, D., "Dependency Theory: A Formalism and Some Observations", 
V.40, December 1964. 

4. Kay, M., "Experiments with a Powerful Parser", RAND, Santa Moniea, 
California, 1967. 

5. Klein, S., "Automatic Paraphrasing in Essay Format", Mechanical 
Translation, 1965. 

6. Lamb, S., 'The Sememic Approach to Structural Semantics", American 
Anthropologist 1964. 

7. Schank, R., "A Conceptual Dependency Representation for a Computer- 
Oriented Semantics", Ph.D. Thesis University of Texas, Austin 
1969 (Also available as Stanford AI Memo 83, Stanford Arti- 
ficial Intelligence Project, Computer Science Department, 
Stanford University, Stanford, California.) 

8. Schank, R., "A Notion of Linguistic Concept: A Prelude to Mechanical 
Translation", Stanford AI Memo 7~. December 1968. 

9. Sgall, P., "Generation, Production and Translation", Presented to 
1965 International Conference on Computational Linguistics, 
New York. 

IO. Thorne, J., Bratley, P., and Dewar, H., '~he Syntactic Analysis of 
English by Machine", in Machine Intelli~ence III, University 
of Edinburgh, 1968. 

II. Walker, D., Chapin, P., Gels, M., and Gross, L., 'Recent Develop- 
ments in the MITRE Syntactic Analysis Procedure", MITRE Corp., 
Bedford, Mass., June i~6. 
