I-I 
FEASIBILITY 
This paper discusses the feasibility of applying a 
model of language use based on a modification and extension 
(to be discussed below) of the generative semantic 
(transformational) theory of language competence recently 
developed by Paul Postal, George Lakoff, John Robert Ross, 
~ames D. McCawley, and others, to problems of computation- 
al linguistics. 
The theory of generative semantics, to be discussed 
in section II, is an outgrowth of, and reaction to, 
Chomsky's 1965 theory of transformational linguistics. 
It is a radical theory which deals with a very great 
range of problems with very abstract methods. Trose 
working in this paradigm hold that there is a linguist- 
ic level reflecting conceptual or semantic structure 
which is directly convertible into surface syntax by a 
single set of garden-variety transformations, with no O 
significant intermediary level, that is, no deep 
structure". These of us working in generative semantics 
believe that methods substantially those long familiar 
in linguistics can achieve very absract , very general 
results which treat semantics in a more serious and en- 
lightening way than ever before. I do not, I think, 
support this very strong claim very well in section II, 
but I provide summaries of several studies and a lengthy 
bibliogrpahy of works which when consulted will hopefully 
give some feeling for what is being attempted, I think 
not without results. 
But generative semantics is a model, or rather, a 
theory, of competence, like most serious theories of 
language now held to by American linguists. ~ven if, 
as might be claimed, our semantic structures are to be 
merely variants of the structures long familiar from 
formal logic, so that if our assumptions are correct, 
we will ultimately be able to directly transform 
surface structures into underlying semantic structures, 
the majority of actual sentences, as well as all hyper- 
sentential structures, the treatment of which has been 
swept under the rug of "performance", will remain unhandle- 
able. 
Accordingly, I propose initially cert@in extensions 
and modifications of the theory to make it in some sense 
1-2 
a model of performance. But if we are to apply it to the 
computer, a major component must still be added. The impe- 
tus to this application is ~he possibility of creating 
an understanding machine, dewcribed in section IV below. 
Since the actual human interpretation of language depends 
on past knowledge (consider which of these sentences is 
good and why: 
As for Albuquerque, the ~iffel Tower is pretty. 
As for Paris, the Eiffel ~ower is pretty. 
And the se : 
ShirLey is a blonde and Susan is Nordic-looking too. 
Shirley is a linguist and Susan is Nordic-lloking too.) 
the old split between semantics, syntax, and pragmatics 
must be revised, and our model closely linked with a 
memory and possibly a logic component as well. 
Obviously this defines a very difficult task, but insofar 
as such goals as HT, artificial intelligence, and machine 
reading of handwritten material or writing of spoken 
material involve comprehension on the part of the machinej 
o~ which there seems to be no doubt, these important 
goals will continue to ~lude us until such time as we 
can devise such an understanding machine as I have ~escribed 
below. 
I believe that generative semantics lays the 
foundation for studies relevant to such a development, 
and it is in this context that my proposals are made. 
In section II I will d~scuss generative semantics. 
In section III I will discuss the body of my proposals 
here. 
In section IV I will discuss what should be required 
of a generalized "understanding" machine. 
.... tl I II \] \] ...... I \]\] 
Part II. The theory of 
Generative Semantics. 
~ae theory of ~enerative semantics is an out- 
growth and reaction to the theory of transformational 
grammar as represented in Chomsky's 1965 book, As s~cts 
of the Theor~ of S~tax (MIT Press). To a very ~-I-~ 
extent, this theory has been the development of a 
small group of former students of Chomsky,s or their 
close colleagues. John (HaJ) Ross has said that the 
theory is really Just an attempt to explicate Pa~l 
Postal's work of five years ago to date. If Postal 
was the founder of this school, if you can call it that, 
its main workers have been HaJ Ross and George Lakoff, 
who between 1965 and 1968 swept aside most of transforma- 
tional linguistics as it then was. But perhaps best 
known of the group is J~mes McCawley, who graduated 
from MIT in 1965 with a Ph.D. based on work in ubono- logy, not syntax or semantics. He promptly amazed Lakoff and 2oss by 
some very substantive work in the latter areas as well as phonology. 
s~udent of McCawley's I will be emphasizing his 
contributions here, and those of my collea~s at 
Chicago, Jerry L. Morgan and Georgia M. Green, but 
it should be kept in mind that people like Ross, 
Lakoff, Postal, Arnold Zwicky, David Perlmutter, 
Emmon Bach, Robin Lakoff, and several others, have 
made the current theory possible, and that many others, 
such as Robert Wall, Lauri Kartunnen, Ronald Langacker, 
and others, have contributed as well. It should also 
be kept in mind that the Case Grammar of Fillmore 
and the work done by Gruber, while differing from 
generative semantics, have contributed a great deal 
to it. 
~ae basic theory of generative semantics is 
built upon an attempt to relate the underlying 
semantic structure of language to the surface, phonetic 
manifestation of that underlying structure. That is, 
a phonetic reality is recognized, and a semantic 
reality is recognized. But unlike other versions of 
transformational grammar, this theory assigns no special 
status to syntax; syntax is subsumed in the semantics. 
McCawley has Jokingly referred to his theor~ as being 
one of either "semantax" or "synantics". 
~11e name generative semantics is not a particularly 
good one, since it implies that the ~oal of the theory 
is, as with the work of Chomsky, to "separate the 
grsumuatical sequences" of a language"from the ~E~__ammatical 
Sequences." (Chomsky, S~_~ctic Structures~ I~.) In 
S 
As a 
II-2 
other words, to generate all and only grammatical sent- 
ences of a language. ~his is not at all the goal of 
generative semantics. Rather, what we want to do is 
in some rigorous way specify the correlations of under- 
lying semantic entities and surface phonetic entities: 
to specify for any underlying semantic structure what 
its possible phonetic realizations in some language are, 
and for some phonetic structure what underlying semantic 
structures it can represent. Naturally, so~e descript- 
ive ability is predicated as well, that is, we want to 
be able to define ambiguity in some algorithmic fashion, 
we want to be able to define levels or classes of 
ill-correlation between structures on different levels, 
etc. Chomsky would say that a sentence~like "Golf plays 
John" is eminently deserving of a star; we would say 
(I) if it's supposed to mean'John plays golf', it doesn't 
succeed in conveying the message; (2) if it's suppesed 
to mean 'John loves Marsha', then it's really bad; and 
(3) if Golf is a man.s name and Gohn the name of a 
game or role, it's a good sentence --- indeed, one 
can very well imagine arcane circumstances under which 
one might utter that sentence with the intent of 
saying that the game plays John, that the tail wags the 
dog~ as it were. Suppose, for example, that John's 
wife were tired of him spending all his free time 
playing golf and she grumbled to a heighbor about 
it, and the neighbor rather unfeelingly replied, 
"Oh well, John plays golf." I can ~ery well imagine 
John's wife complaining bitterly, "Oh no, golf plays 
John." In any case, it is for hus unimaginative 
approach to language that Chomsky has been Jokingly 
called a "bourgeois formalist"o Even when we use 
stars, we try to keep in mind that Just about 
any valid phonological string of a language conveys 
one or more meanings in some context, and that it is 
artificial to take a string out of context and declare 
it good or bad. So "generative semantics" is a bad 
ns~e. 
The following diagram of the components of the 
theory is based on McCawley's paper in the proceedings 
of the 4th Regional Feeting of the Chicago Linguistic 
Society (1968). A theory very similar is discussed 
in Ronald Langacker's book LAnguage and its Structure 
(Harbrace, 1968), pp. 114-34. 
II-3 
WHATEVER 
INITIATES 
THO UG HT 
semantic ITRANSFORMA- | surface 
-->represent- --~TIONAL ~repre- 
ation |COMPONENT |senta- 
..... tion $ 
phonetic . |PH0 NOLOGIC- 
represent a-~--~AL 
tion ~COMPONENT_ 
The above diagram comes from a report prepared by myself, 
Jerry Morgan, and Georgia Green, called the Uamelot 
~o' which attempted to describe the curren-~te of 
rmational re, search in the Sum~uer cf 1968, particu- 
larly in reference to the LSA Summer Linguistic Institute 
at the University of lllinois, where HaJ Ross, George 
Lakcff, and Jim McCawley had lectured to large groups 
on a huge number of very '~airy" (i.e., difficult and 
tickleishly novel) topics. 
In that report (which was prepared for Victor Yngve ), 
we raised several questions concerning the above repre- 
sentation. We asked: 
i. ~hat will an adequate semantic representation have to 
include? What form will it have? 
2. ~hat can a transformation do? What does one lock 
like ? 
stage 
3. At what A and in what manner are semantic 
repres@~tations converted into words of real 
languages? 
~ese were by no means all of the questions asked. 
Needless to say, the answering cf these questions 
has hardly begun and will undoubtedly guarantee 
linguists a few gocd centuries of work at least. It 
is only in the last decade that syntax has been the 
subject of serious work, and we are still only 
discovering how ignorant we are. Semantics is even 
newer, less than a decade old. If anyone doubts 
that this is true, consider a) what the above 3 
questions would have meant to a linguist in (say) 
1955, and b)why he would have been wrong in his 
(lack of) comprehension of them. One of the 
great contributions of Postal and Ross has been 
II-4 
their constant critical look at transformational grammar. 
One of the things they saw was that our transformations 
were (and are) extremely powerful devices, with practical- 
ly no constraints placed on their formulation. 
~at I will do here is summarize some of the 
attempts at partial answers to the three above questions. 
In this way I can delimit and explicate generative seman- 
tics best. 
i will start by abstracting parts of two papers 
by McCawley that deal with the nature of semantic representa- 
tion. In a paper in the Japanese Journal ~otoba no Uchu 
(World of Language) in 1967, McCawley argued tha~ semantic 
representation would be similar to syntactic representation 
as familiar from ~-type grammar, but that it would 
also be quite similar to symbolic logic as familiar from 
the tons of work that have followed Principia and such 
studies. That semantic representation should resemble 
syntactic representation makes sense if only because 
we are arguing for a single set of rules that transforms 
(i.e., reEates) the underlying structure into (to) 
the surface structures. There will be more about that 
later. 
McCawley argues as follows: the following devices 
have all had a role in symbolic logic: 
I. propositional connectives" 'and', 'or', 'not'. 
2. constants denoting individuals. 
3. predicates, denoting properties and relationships. 
4. set symbols and the quantifiers 'all' and 'there 
exists '. 
5. descriptions of sets and individuals. 
• x The following devices play a role in natural languages : 
I. all igs. have words for 'and', 'or', and 'not'. 
(he notes however that these words in natural igs. 
may connect more than sentences ) 
2. "indices" denoting individuals; John loves John 
might be represented as x I loves x2, but John loves 
II-5 
himself is x I loves x I. 
3. predicates are expressed in natural Igs. (by verbs, 
adjectives, nouns, etc.) 
expressioms such as 
4. "Words such as all and&at least one are two members 
of a rather larg--e--clasB of expressions which are 
used to indicate not only the existence of an 
individual or a set but the absolute or relative 
number of members in that set." 
5. sets and individuals can be expressed as descriptions 
using modified noun phrases. 
McCawley then gives further reasons for supposing symbolic 
logic representation to be proper for semantic representa- 
tion. (See the bibliography to this section where t~is 
and other papers that can be consulted for these arguments 
in detail are listed. ) 
In a paper prepared for the symposium on "Cognitive 
Studies and Artificial Intelligence Research" held by the 
Wenner-~ren Foundation at the University of Chicago in 
March of this year, McCaWley discussed semantic representa- 
tion at length. Some of what he had to say there should 
be noted. He claimed, "semantic representation must indicate 
the immediate constituent structure of the elements invol~ed 
in it {i.e. examples showing that different meanings can 
comsist of the same semantic elements combined in different 
ways\[ are easy to come b~)" {p.l) He gave the example of 
John doesn't beat his wife because he loves her. 
If the negation applies to John beats his wife, the 
se'~tence means 'the reason ~at John doesn't beat his wi~e 
is that he loves her', whereas if it applies to the 
John heats his wife because he loves her., the mg. is 'the 
reason that John beats his wi~e -~not- ~Hat he loves her. ' 
Notice that here a surface form represents at least two 
different underlying structures which nonetheless contain 
precisely the same semantic elements-- grouped differently, 
however. 
Another point made is that "semantic representations 
must include .., some indication of presupposed coref- 
erence." (p.2) That is, the followingsentence in neutral 
(i.e. null) context is ambiguous three-ways: 
II-6 
John told Harry that his wife was pretty. 
Whose wife? John's? Harry's? or a third's? It could be 
any. However, if we know who his refers to, there 
is no such ambiguity. This may--seem trivial, but 
it is a point often ignored. 
McCawley then gives an argument for referential 
indices being different from expressions used to de~ribe. 
The sentence 
Max d~bied that he kissed the girl Be kissed. / 
is not contradictory if "the girl he kissed" is the 
speaker's description. 
Another notion is that of presupposed set membero 
ship. lh 
Max is more intelligent than most Americans. 
said with primary stress ~n most, the sentence is good 
if and only if Max is presupp--~a to be American, that 
is, the sentence implies Max is American. With primary 
stress on Americans, however, Max is presupposed not 
to be American. Presupposition is in general a very 
hairy topic which was recently the subject of an entire 
conference (at the Ohio State University). We know 
very little about the nuamces of implication and 
are only beginning even to identify the problems. But 
if a machine is ever to rea~ Ga_tcher i~ the Rye catching 
all the nuances of the italicized words, we had better 
find out how stress is used to alter the presuppositlon- 
al set of a sentence. I need not be so unsubtle as to 
suggest the extreme value of such researches to psycholo- 
gy. Perhaps they already know about all this, for all 
I know. In any case I cannot restrain myself from 
inclucing McCawley's beautiful example 
CIA Agents are more stupid than most 
Americans. 
He had primary streos on the ~ but I prefer to think 
of it as going on the Americans. 
Z would like ~o interject at this point a minor 
apology. I have been rather fan-clubish here and 
have waved my hand a lot. Frankly I see no value 
8 
11-7 
in rehearsing here all the arguments available elsewhere. 
But I would like the rea~r to bear in mind my skimpy 
resume in no way reflects~the quality of the original. 
Let me also note, lest I seem u~uduly credulous towards 
tjheDthoughts of. C~irman Quang 4mild-maunered linguist 
• . mc~awley Is In reality Q. p. Dong, Chairman of 
Unamerican Studies at an unknown universityJ, that most 
of us working within the paradigm of generative 
semantics would be the first to admit that our theories 
haven't a pra~er of being right, that is, they~~ 
approach even a partially realistic and naturalistic "J 
theory of language. If we like it better than other 
paradigms it is because we believe that no 
other cureent theory is any better and that this one 
at least has a good chance of self-improvement. (End 
of apologia. ) 
If semantic representation looks much like 
logical representation, it also differs from it. 
In the Kotoba no Uchu paper McCawley noted the follow- 
ing differences : 
I. "It is necessary to admit predicates which assert 
properties not only of individuals but also of 
sets and propositions." 
2. "In mathematics one enumerates certain objects 
which ~one~will talk about, defines other obJecSs 
in te~ms o? these objects, and co~Ifines\[onesel~ 
to a discussion of objects which\[oneS has either 
postulated or defined .... However, one does not 
begin a conversation by giving a list of postu- 
lates and definitions..... •..people often 
talk about things which either do not exist or 
which they have identified incorrectly• indices 
exist in the minds of the speaker rather than 
in the real world; they are conceptual entities 
which the individual speaker creates in interpret- 
ing his experience." 
In the Wenner-Gren symposium, McCawley had more to say about 
the difference between logic and language. 
1. Immediate constituent structure (trees)rather than 
parentheses are basic. First, "semantic representations 
are to form the input to a system of ~ransformations 
that relate meaning to superficial form; to the 
II-8 
extent that these transformations have been formulated 
and Justified, they appear to be stateable only in 
terms of constituent structure and constituent type, 
rather than in terms of configurations of parentheses 
and terminal symbols." Secondly, "it may be necessa- 
ry to operate in terms of semantic representations in 
which symbols have no left-to-right ordering .... " 
2. There will have to be more 'logi~al operators', such as 
most, almost all, and m~. 
3. "And and ... or ... cannot be regarded as Just binary 
operators but-~ust be allowed to take an arbitrary 
number of operands." 
4. The quantifiers must be restricted rather than 
unrestricted as in most logical systems. Some 
quantiflers imply existence: All dogs like to 
bite postmen, involves the presupposition that 
dogs exist, whereas the unrestricted quantifiers 
logicians use have no such presupposition. 
5. "Adequate semantic representation of sentences 
involving'shifters' (Jakobson, 1957) such as 
I, YOu~_ ~ now, ..., gestures and deictic 
~6rds like this--and that, and tenses, will have 
to include re~rence-K~-the speech act. The most 
promising approach to this aspect of semantic 
representation ... is Rose's (1969)elaboration 
of Austin's (1962) notion of 'performative verb'." 
(See now too Searle's book, Speech A~ts, CUP, 1969 
---RIB) 
6. "The range of indices will ~ave to be enormous. 
In particular, it will have to include not only 
indices that purport to refer to physical objects, 
but also indices corresponding to mythical or 
literary objects, so that one can represent the 
meaning of sentences such as 
The Trobriand Islanders believe in Santa 
61aus, but they call him Ubu Ubu." 
7. McC. rejects "the traditional distinction between 
'predicate' and 'logical operator' and trea~s~ 
such 'logical operators' as quantifiers, conjunctions, 
and negation as predicates...." 
10 
II-9 
To clarify the relationship of semantic to syntact- 
ic representations let me quote here from McCawley's 
Kotoba no Uchu paper: 
Since the rules for combining items into larger 
units in symbolic logic formulas must be stated in terms 
of categories such as 'preposition', Ipredicatel, and 
'index'~ these categories can be regarded as labels on 
the nodes of these trees. And since ... these categories 
all appear to correspond to syntactic categories, the 
same symbols (S, V, NP, etc. ) may be used as node labels 
in semantic representations as are used in syntactic re- 
presentations. Accordingly, semantic representations 
appear to be extremely close in formal nature to syntactic 
representations, so close in fact that it becomes possible 
to catalogue the conceivable formal differences and determine 
whether those differences are real or apparent° 
Among such differences he lists: 
I. "The items in a s#ntactic representation must be 
assigned a linear order, whereas it is not obvious 
that linear ordering of items in a semantic repre- 
sentation makes shy sense." 
2. "Syntactic representations inwolve lexical items from 
the language as their terminal nodes, whereas the 
terminal nodes in a semantic representation are 
semantic units rather than lexical units." 
"There are many syntactic categories which appear to 
play no role in semantic representation, for ex., 
verb-phrase, preposition, and prepositional phrase." 
(At the 5th Regional Meeting of the CLS, April of 
this year, A. L. Becket of the University of Michigan 
presented a paper in which he argued prepositions are 
underlying predicates; prepositional phrases are 
accordingly verb-phrases. ) 
McCawEey concluded nonetheless that these differences 
do not provide an argument that semantic ~epresentations 
are different in formal nature from syntactic representa- 
tions. Again, I will omit his reasons for that conclusion. 
I might summarize all this by saying: 
i. Semantic representatio~ is a modification of the 
representations long familiar from ~ormal logic. 
B. 
11 
II-lO 
2. Such representations do not radically differ from 
the surface syntactic representations of Aspects- 
type grammar. 
Let me close by posing more problems. McCawley asks 
the following questions at the end of his Eotoba no 
Uchu paper. While they do not specifically reEate 
to semantic structure, I include them to give some idea 
of what we believe to be the sort of questions that 
a serious theory of language should prowide Justifiable 
answers for : 
I. How do the mgs. of words change as a language evolves? 
2. How does a child learn rags. in learning to speak his 
native language? 
3. ~at mechanisms are involved in phenomena such as 
metaphor .... ? (Dorothy Lambert has written a 
Ph.D. thesis at Michigan on the subject of 
metaphor within the paradigm of Case Grammar. 
This 500 page dissertation is probably one of 
the best studies of the subject to date from a 
linguistic point of view.S--RIB) 
4. To what extent are the units of semantic representatiomq 
univers el ? 
5. To what extent does the lexicon of a language have 
a structure? 
6. Can all languages express the same ideas? 
7. To what extent doe's one's language affect his thinking? 
8. To what extent is one's ability to learn lexical items 
conditioned by his knowledge of the world? 
I will now turn to the second question raised 
above on p. II-3o This question has as yet received 
little study. It is a very difficult topic, but a 
very important one. I will confine myself here to a 
few brief comments and a few references. 
12 
II -Ii 
One of the important studies underway now is about 
syntactic variables. This was the subject of Ross' 1967 
dissertation. Variables such as X and Y are familiar 
from transformational grammars, but no one had attempted 
before to specify in general what the notion of syntactic 
variable entailed. While Ross' study was important, and 
he came up with several important constraints on the 
form of transformations, much work remains. Lakoff an~ 
Postal are also working on related questions. Let me 
llst here some of the constraints Ross gave in his thesis: 
I) The complex NP constraint. 
No element contained in a sentence dominated 
by a mounphrase withxa lexical head noun 
may be moved out of that noun phrase by a 
transformation. (p, 127) 
2) The o~oss-over condition. 
No NPmentioned in the structural index of 
a transformation may be reordered by that 
rule in such a way as to cross over a 
coreferential NP. (p. 132) 
3)T~e coordinate structure constraint. 
In a coordinate structure, no conjunct may be 
moved, nor may any element ~ontained in a 
conjunct be moved out of that conjunct. 
(p. 161 ) 
~) The pied piping convention. 
Any transformation which is stated in such a way 
as to effect the reordering of some specified 
node NP, where this node is preceded and followed 
by variables in the structural index of the 
rule, may apply to this NP or to any non- 
coordinate NP which dominates it, as long as 
there are no occurences of any coordinate node, 
nor of the node S, on the branch connecting 
the higher node and the specified node. 
(That is,: 
...any NP above some specified one may be reorder- 
e~, instead of the specified one, but there are 
environments where the lower NP ~ay not be moved, 
and only some higher one can, consonant with the 
conditions imposed ~rn the convention.~7) (p.206) 
i3 
11-12 
5) The sen~ential subject constraint. 
No element dominated b~ an S ~ay be moved 
out of that S if that node S is dominated 
hy an NP which itself is immediately 
dominated by S. (p. 243) 
6) The frozen structure constraint. 
If a clause has been, extraposed from a noun 
phrase whose head noun is lexical, this noun 
phrase may not be moved, nor may any element 
of the clause be moved out of that clause. (p. 295) 
7) Definition of identity. 
Constituents are identical if they have the 
same constituent structure and are identical 
morpheme-for-morpheme, or if they differ only 
as to pronouns, where the pronouns in each of 
the identical constituents are commanded by 
antecedents-in the non-identical portions 
of the phrase-marker. (p.348) 
A very important constraint occurs on p. 480 of the 
thesis, but I omit it here because it contains many 
terms I would not care to define here. I reccomend 
Ross' dissertation for anyone with doubts about 
any deep principles of language organization emerging 
from our studies in transformational grammar. He will 
be cured. 
Recently George Lakoff has studied the notion of 
"derivational constraint". This study is quite recent 
and still very very hairy, but hints in his 1969 CLS 
paper, and comments by Postal on it suggest that rule 
odering is merely a special case or manifestation 
of a deeper principle of grammar organization. The 
next revolution effected by generative semantics may 
well be to drop rule ordering from our canons. 
For various reasons (partly that it interests 
me mere ) I will have much more to say here about 
lexlcal insertion than I will about constraints on 
transformations, although undoubtedly the . latter 
is ultimately o~ much greater importance. 
Until 1965 or so, it was assumed that the terminal 
symbols of a P-marker are lexical items; the lexicon merely 
assigns properties to these items. Bruber in his 1965 
14 
If-13 
dissertation argued that certain transformations had to 
occur before lexical items entered trees: that is, that 
there were pre-lex±cal transformations. 
Before Gruber, the system of semantics was one 
in which T-rules generated from deep structures surface 
str%~ctures and P-rules generated semantic representations 
for those deep structures. • T~lis was the theory of 
intepretive semantics (as in Katz and Postal, for ex. ) 
Gr~ber proposed a derivational semantics. Gruber 
intended to "show va-~ consistently recurrent semantic 
relationships among parts of the sentence and among 
different sentences, which can best be explained by the 
existence of some underlying pattern of which the 
syntactic structure is a particular manifestation." (p.l) 
He concluded that "a level at which semantic interpretation 
w~ll be relevant will ... be deeper than the level 
of 'deep structure' in syntax." (p.2) Later Lakoff showed 
evidence that in fact the level of semantic interpretation 
was that of deep structure, but argued that (as Gruber 
said) "syntax and semantics will have the same representa- 
tion at the prelexical level"(p. 3): a single set of rules 
would transform semantic structures containing no lexical 
items into surface syntactic representations containing 
them. 
The s~udy of lexical insertion, the process by 
which the underlying semantic elements are grouped into 
units replaceable ~y surface lexical items has led to 
a large literature containing a great many questions, 
and some positive answers. An important paper was 
McCawley's 1968 paper, "Lexical insertion in a transforma- 
tional grammar without deep structure." 
There he started by assuming various points 
concluded in other papers of his. He very clearly 
presents some of the tehots of generative semantics, 
so with some repetition from above I quote these 
points here : 
I. Syntactic and semantic representations are of the 
same formal nature.... 
2. There is a single system of rules ... which relates 
semantic representation ~o surface structure through 
intermediate stages. 
3. In the earlier stages of the conversion from semantic 
15 
11-14 
representation to surface structure, terminal nodea 
may have for labels 'referential indices' such as 
were ~ntroduced in Chomsky 1965 .... In semantic 
representation, only indices and 'predicates' a~e 
terminal node labels .... 
McCawley then defined tdictionary entry' as a 
transformation which replaced part of a tree by a 
surface lexical item. He expressed doubt these rules 
could be ordered internally or external, since it 
would hardly be possible, for example, that some question 
would arise as to the relative ordering of the transforma- 
tion introducing the word horse and that extraposing NP's 
in two dialects, that is, ~he ~rdering could not possibly 
matter. 
He then raised several possibilities as to the re- 
lative ordering of the lexical rules v~s-a-vis other 
rules. Are the lexical rules last, first, or where? 
McCawley argued for the lexical rules applying Just 
before the post-cyclic rules, and adduced evidence for 
several rules, predicate-raising, equi-NP deletion, etc., 
being pre-lexlcal. 
In his 1968 LSA paper, Jerry L. Morgan of the 
U;~Iversity of Chicago added to this. He pointed out 
'the rather strong assumptlon that lexical_items only 
'replace' constituents." (P.3) He wrote, "~he process 
of syntactic derivation begins with semantic representa- 
tion in terms of trees containing very highly abstract 
semantic terms, operating upon this by means of rules 
permuting, deleting, and collapsing parts of the representa- 
tion, finally deriving a structure whose constituents are 
replaced by lexical items." (p. 3) He then s~ated a 
very strong claim of the theory: 
Given the set of universal pre-~exical 
rules, the set of universal semantlo 
primitives, and the set of universal 
constraints on the operation of rules, 
such as those described by Ross 1967, 
these define the universal set of possible 
lexical items in their semantic aspect; 
that is, they rule out as impossible am 
infinit~ classof a priori possible 
"meanings" a lexlcal item could have. (P.4) 
16 
If-19 
A second very strong claim of the theory is: 
Insofar as the selection from, and details 
of implementation of, the universal set of 
rules is language-specific, the idiosyncracies 
of a given language in this respect will also 
be reflected by systematic gaps in the lexicon. 
The same is true for the set of semantic 
primitives and the se~ of constraints on rules. 
.... (P.4-5) 
Morgan came up with some restrictions on lexical items : 
only 
I) "lexical items Jan replace a constituent which ,! 
is not labelled S. (p.6) 
2) "verbs cannot incorporate referential indi~es."(p.6) 
One l~Lher point to be made is that lexical items 
can only replace well-formed sub~rees. 
My own work has been concerned with specifying 
classes of possible lexical items and accounting for 
the syntactic properties of verbs in terms of their 
semantics, thereby attempting to capture the intuition 
long familiar from traditional grammar that certain 
~emantic classes of verbs, such as "verbs of giving 
and taking" or "verbs of motion" also form syntactic classes 
and hence their syntactic properties can be regarded as 
derived from their semantics. 
Georgia Green of the University of Chicago has 
presented a paper (1969) which is also interesting 
in terms of lexical insertion. She tends to regard 
lexical insertion hs fairly ~ivorced from 
morphology, and views lexical insertion as the replace- 
ment of an entire sub-tree by a surface lexical item 
which may contain more than one morpheme as ~lassically 
defined. This position is somewhat different from my 
own, as I regard lexical insertion as primarily involving 
the replacement of items on a i-I basis. However, this 
is an empirical question and only future research will 
decide which of us is more nearly correct. 
So far I have discussed lexical ~nsertion in 
terms of sweeping, general p~inciples of the organization 
of the grammar. I~ order to more clearly specify what 
17 
II-16 
lexical insertion is all about, I ought to present 
some of the kinds of problems which have generated 
811 of this interest in the subject. 
f 
At the Texas Conference on Universals in 1967, the proceedings of whioh were 
p~lished in 1968 as Universal_s in Linguistic Theory, McCawley raised the 
~e~0sals in L, nguistic Theory, McCawley ralsed the 
ques~o~ dictionary organiza'tion anew. He opted fo~ 
a "Weinreichian" lexicon in which lexical items were 
combinations of semantic, syntactic~ and phonological 
information. McCawley supported this with this evidence: 
the reason John is sadder than that book. is bad is that 
the two sads in the underlying structure of the sentence 
are d~erent lexical items. They therefore cannot 
participate in comparison: 
*John is as sad as that book he read yesterday. 
*He exploits his employees more than the oppur- 
tunity to please. 
*Is Brazil as independent as the continuum 
hypothesis? (exx. of Chomsky's. ) 
McCawley called for a theory of "implioaticnal relations", 
since in cases such as the ambiguity of warm the ambiguity 
is not a property of the item itself but-B'~--a class of 
items, and therefore such an ambiguity must be specified 
in terms of general principles. NcCawley was not clear 
about the nature of these implicational r~lations, so 
that the nature of the relationship of the various sads 
was more or less left open. I have discussed the no-~on 
of systematic ambiguity, where the ambiguities of an entire 
class of verbs is specified in terms of the derivational 
process underlying them all, not Just in terms 
o~ a descriptive statement. Thus we are seeking to 
explain lexical gaps in terms of statements such as 
"The reason some language L lacks a verb ¥ 
glossing the verb W in the language M is that 
M, but not L, has the transformation T." 
Anyone familiar with the lexicons of French, ~n~lish, 
and German, for example, knows that there are certain kinds 
of verb which are not typical of one or another of these 
languages which nonetheless readily occur in the others. 
Such verbs are derived by processes occuring in one but 
not another language, and our task is to discover and 
describe such processes. Thus we may ultimately be able 
to tell how the class o~ French verbs, say, differs from 
18 
II-17 
the class of all possible verbs. 
I have attempted in these few pages to present 
a digest of some works in the paradigm of generative 
semantics. I have not really attempted to provide 
even an elementary guide to the methods of generative 
s~mantics or to its conclusions, its findings, but I 
hope I have explicated somewhat its goals and given 
some insight into the direction in which it is moving. 
Some very strong claims are forthcoming on the nature of 
grammars and languages and hence of language itself. A 
tremendous amount of work needs to be done, but one can 
see clearly that one possible end point of this work will 
be a very comprehensive, very strong theory of language 
competence that has a great deal to say about human be ings. 
One perhaps minor point, though, looms up large 
here: generative semantics relates semantic structures 
to stu~face sentences by a single Eet of r~les. There 
are s-~veral versions of transformational grammar that do 
this, but generative semantics is perhaps the most-Cevelop- 
ed of these. But as the saying goes, what goes up must 
come down: we may paraphrase this as: what can be generated, 
can be analyzed. T~e theory permits, idsally, an 
algorithmic translation of a surface string into one or 
more underlying semantic structures. For computational 
linguistics, that may be its most appealing feature. 
19 
II-18  BIBLIOGRAPHY 
A short, select blblio~raphy of recent works in and 
on gener~ ~ive semantics. 
Austin, J. L. 
1962. How to do things with words. 
London: 0UP. (1965 paper.) 
ed. J.0. Urmson. 
Bach, Emmon. 
1964. "Have and be in English syntax." Lg. ~3.462-85. 
196~. "Problomlnalization I-II." Mimeo. 
196~. "Nouns and nounphrases." in Bach & Harms. 
1969. "Anti-pronomlnalization." Mimeo. 
Forthcoming. "Binding." 
Bach, E~on, and Peters, Stanley. 
1968. "Pseudd-cleft sentences." Mimeo. 
Bach, E~m~n, and Harms, Robert, edd. 1968. Universals in Linguistic Theory. NYC: Halt,Rinehart, 
Winston. 
Becket, A. L. 
1969."Prepositions as Predicates."in Binnick et al. 
Benwick, Launcelot de, the Green Knight, and Morgan le Faye. 
(:R. Binnick, G. M. @teen, J. L. Morgan. ) 
1968. Camelot 1968. Internal memo., MT Group, UC, mimeo. 
Bierwlsch, Manfred, and Heidolph (edd) 
to appear. Recent Advances in Linguistics. l~uton. 
Binnick, 
1967. 
1967. 
1968. 
1968. 
1968. 
1968. 
1969. 
Robert I. "Semantic and syntactic classes of verbs." Mimeo. 
"The lexicon in a derivational semantic theory 
~ofJrna~Sf~i~°~ lslnlg~i~tics 1 In Chicago i i ab from Univer- 
sity Microfilms, AnnA~bor, Michigan, as ser~l 
s-372. "On the nature of the 'lexical item'", 
in Darden etal. "On transforma~ionally derived verbs in a 
~rammar of English", .Ditto, read at LSAo The characterization of abstract lexical 
entities", ~Ditto, read at ACL. "Transitive verbs and lexical insertion", 
dittoed, read at Kansas and CLS. "Predicative structure." Unpublished Ph.D. diss. 
20 
II-19 
Bin_nick, R., G. Green, J. Morgan, & A. Davison, edd, 
1969. Papers from the 5th Regional Meeting, 
Chicago Linguistic Society. Chicag6: 
Department of Linguistics, University of 
Chicago. (Advt. : dirt cheap at $5. ) 
Camelot. see Benwick. 
J~arden, B. J., C. J. Bailey, A. Davison, odd. 
1968. Papers from the 4th Regional Meeting, 
Chicago Linguistic Society. Chicago: 
Department of Linguistics, University of 
Chicago. (Advt. : still dirt cheap at $3. ) 
OeRiJk~ Rudolph. 
19~J. "A note on prelexical predicate raising." 
Dittoed. 
Donnellen, Keith. 
1966. "Reference and definite descriptions." 
Philosophical Review 75. 281-304. 
Green, Georgia M. 
1968."0n too and either, and not Just too 
an~ eith-e-r, either-~-7-in Darden et a1~---- 
1969. "Some---~-eoretical implications of the lexical 
expression of emphatic conjunction." Unpublisheo M.A. thesis, dittoed. 
1969. "On the notion 'related lexical entry. '" in Binnick et al. 
~orthcoming. Review of R. Lakoff, Abstract Syntax 
and Latin Complementation. To appear in L~. 
Gruber, Jeffrey S. 
1965. Studies in lexical relations. Unpublishe~ diss. 
1967. "Look and se~', in L~K. 43.9~7-47. 
1967. "The functions of t~-6 lexicon in formal 
descriptive grammars." Systems Development 
Corporation doctuuent TM-3770/000/00. 
Jakobovits and Steinberg, edd. to appear. Semantics: an interdisciplinary reader .... 
Jakobson, Roman 
1957. "Shifters,verbal categories and the Russian verb." 
Slavic Dept., Harvard Univ. i 
21 
II-20 
Kartunnen. Lauri. 
1968. ~Co-reference and discourse." Read at LSA. 
1968. "~hat do referential indices refer to?"~ RAND 
Corporation Rgport. 
1969. "Migs and pilots. " Mimeo 
1969. "Pronouns and variables. ~' In Binnick et al. 
Katz, J. and Postal, Paul. 
1964. An integrated theory of linguistic descriptions. 
HIT Press. 
Kiparsky, Paul. 
1968. "Linguistic Universals and Linguistic Change." 
in Bach and Harms. 
Kiparsky, P. & C. 
1968. "Fact. " To appear in Bierwisch and Heidolph. 
Lakoff, G~orge. 
1965. "On the nature of syntactic irregularity." 
Indiana Univ. diss. =NSF-16 report, ed. 
A. 0ettinger. Available from : 
Clearing House for Federal Scientific 
and Technical Information, Springfield, Va., 
as document PB 169 252 ($3). See also 
0ettinger below. (Alias NSF#~7). 
1966. "Stative adjectives and verbs in English." 
"A note on negation." 
Both in NSF-17. ~ 
1967. "Pronominalization, negation, and the analysis 
of adverbs." ms. 
k1966. "Deep and surface gray,nat." ms. Also 
several of these papers currently available 
from Linguistics Club, Linguistics Department, 
Indiana University. 
1968."Counterparts." Read at LSA. Mimeo. 
nd. "Pronoun~ and reference", ms. 
1969. "Some semantic considerations in syntax." 
In Binnick and al. 
add: 
1968."Repartee". To appear in Foundations of Language. 
1968. "Instrumenta~ adverbs and the Concept of Deep 
structure." in Foundations of Language. 
1969. "Presuppositions and relative grammaticality." 
Read at LSA in 1968. 
to appear. "On Generative Semantics" in Jakobovits and 
St einberg. 
22 
11-21 
Lakoff, George, and Peters, Stanley. 
1966. "Phrasal conjunction and s~numetric predicates", in NSF-17. 
Lakoff, George, and Ross, John R. 
1966. "A criterion for verb phrase constituency. " 
I n NSF-17. 
Lakoff, Rob-n T. 
1968. Abstract syntax and Latin complementation. MIT Press. 
1968. "Some reasons why there can't be a some-any rifle." 
Read at LSA, mimeo. 
1969. "Syntac%ic arguments for not-transportation." In Binnick et al. 
/ to appear. Review of Grammaire G~n~rale et Raisonnee, 
1660. To appear in L~. 
Langacker, Ronald. 
1968. Language and its Structure. Harbrace 
to appear. "On pronominalization and the chain of 
command", mlmeo 1966, to appear in an anthology by Reibel and Schane to appear. 
>~ Cawley, 
nd. ~ 
1968. 
1966. 
1968, 
1969. 
1969. 
1969. 
add: 
1967. ~ 
1969. 
19£8. 
James Do 
"The annotated respective." F~Imeo. 
"On the role of semantics in a grammar." In 
Bach & Harms. 
Review of Cooper's Set Theory and Syntactic 
Oescription. in Foundations of Language. 
"Concerning the base component of a transforma- 
tional grammar." in Foundations of Lg. 
"A note on multiple negations." mimeo. 
"English as a VS0 Lg." mimeo, read at LSA 1968. 
"Tense an~ time reference in English." Read 
at Ohio State Semantics funfest; to appear in 
Working Papers in Linguistics 4 of 0SU Ling. dept.; 
~eo, 
"Meaning and the description of ig." Kotoba no 
Uchu, Tokyo, nos. 9,10,Ii. 
"Semantic representation." Read at Symposium 
on Cognition etc. 
"On lexical insertion in a ~ransformational gram- 
mar without d~ep structure." In Darden et al. 
Morgan, J. L. 
1968. "Some strange aspects o£ "it"." In Da~den et al. 
1968. "On the notion 'possible lexical item'". Read 
at LSA. Ditto. 
196~. "Irving." Dittoed. 
23 
II-22 
1968. "Three notes on Irving and other matters." Ditto. 
1969. "On arguing about semantics." Ditto, read at SECOL. 
1969. "On the memantic representation of lexical 
items." In Binnick et al. 
0ettinger, A. ed. 
1966. NSF-17. Available from Clearimghouse as 
document PB 173 630 ($3). 
Perlmutter, David. 
1968. ~e two verbs 'begin'." to appear in an 
anthology by Jacohs and Rosenbaum to appear. 
1968. "aeep and Surface Structure Constraints in 
Syntax. " MIT diss. 
Postal, Paul. 
1966. "On so-called 'pronouns, in English." in 
Georgetown University Monography Series 
on Languages... 19, 177-206. 
1968. "Cross-over constraints." ms. 
to appear-a. "On coreferential complement subject 
deletion" in Jakobovits and Steinberg. 
to appear-b. "On the derivation of surface nouns." in 
L~ngui_stic In u q,u~_~ /a new Journa!J 
1969. "On 'remind'", read at Ohio State. 
1969. "Anaphoric Islands." in Binnick et al. 
Quang Fnuc Dong. 
1968. "English sentences without overt grammatical 
subject." Mimeo. 
1968. "A note on conjoined NP.s", mimeo. 
Ross, John R.~ 
196)."A p~ogsed rule of tree pruning." 
"Relativization in gxtraposed Clauses." 
_~ Both in NSF-17. 
~1967. "Auxiliaries as main verbs." ~.z~c~ , 
~\1966. "On the cyclic nattu~e of English pronominalization. 
\ in Jakobson Festschrift, Mouton. 
~1966. "Adjectives as NP's". 5,~, 
~ 1967. "Gapping and the order of constituents." PEGS. 1968. "O n declarative sentences." 0;~%~ 1967. "Constraints on variables in syntax. " Dies. 
1969~ several forthcoming papers. 
24 
II-23 
Vendler, Zeno. 
19o7. Linguistics in Philosophy. Cornell Univ. press. 
Wall, Robert. 
1967. "Selectional restrictions on subjects and 
objects of transitive verbs." 
Mimeo. 
Zwicky. Arnold. 
1968. "Naturalness argt~uents in Syntax." In Darden et al. 
25 
III-i 
EXTENSION AND MODIFICATION O~ 
TH~ THEORY 
Paul Postal, in a 196~ paper, "Underlying and super- 
ficial linguistic structure", seemed to rule out any 
principled approach to the study of performance. But it 
seems clear to me that performance has me, ely been a 
catch-all term used by linguists with a lot of nasty 
facts on their hands they had no way of handling. In 
section III mentioned the treatment of semi-grammatical 
sentences as they used to be called. Now I think we 
should be able to treat so-called sentence fragments 
as being part of language proper. I see no reason, 
once we get over our hang-ups with sharp categorization 
of grammaticality and Judgments of grammaticality in 
null context, why we cannot have a principled treatment 
of sentence fragments. 
Another area ~sually relegated to Never-never land 
is that of the structure of discourse. Obviously the 
sentence pairs 
Harry is a fool. He voted for Richard Nixon. 
Ha voted for Richard Nixon. Harry is a vote. 
are not equivalent. Imagine if we take every other sentence 
on a page, say, the beginning of Matthew 2. The result is 
hardly a well-formed discourse. 
Now the birth of Jesus came about in this way. 
But her husband, Joseph, was an upright man 
and did not wish to disgrace her, and he decided 
to break off the engagement privately. "Joseph, 
descendent of David, do not fear ~o take Mary, 
your wife, to your home, for it is through the 
in/'luence of the holy Spirit that she is to be- 
come a mother." All this happened in fulfill- 
ment of what the Lord said through the prophet .... 
But he did not live with her as a husband until 
she had had a son, and he named the child Jesus. 
"Where is the newly born king of the Jews?" 
To now, it has generally been held that the structure of 
discourse is linear, that is, sentences are strung to- 
gether one after the other and well-formedness is based 
on kow well these sentences string. But the context 
is vital to the form of a sentence. Similarly, whether 
26 
III-2 
two clauses are united or put into separate sentences de- 
pends on context: by context we cannot mean m~rely the 
two sentences on either side of the sentence in question, 
nor can we mean the n sentences to either side. ~is is 
quite as mad as the foll~ of the early 50's that syntax 
was a matter of which words had what probability of 
occurir~ n words to either side of a given word. ~at we 
need is a grammar, a generative grammar, a transforma- 
tional grammar, of discourse, based on the same methods 
that have been developed in syntax over the last decade. 
This worm was pioneered by George Lakoff's 196~ study 
of Russian folk-tales, in which he revised Propp's 
phrase structure grammar of the "morphology" of Ruszian 
~olk-tales. I subsequently re-modified Lakoff-s work 
and programmed it in COMIT for a 7090-7094 machine to 
generate plot outlines of Russian folktales. The results 
were partly abominable and partly amusing, but the point is 
that while hardly any discourse is as steretyped as 
Russian folktales or US patents, that certain structures 
nonetheless occur which are larger than the sentence. The 
notions of subordination and coordination of sentences and 
even whole discourses are quite valid and quite amenable 
to inve st~gation. 
A third class of problems concern logic. The implica- 
tions of a sentence may be quite as important as the state- 
ments made by it. We linguists are only beginning to 
investigate presupposition, implication, insinuation, 
assertion, etc., but philosophershave been aware of these 
problems for a long time ano a large literature exists. 
We want a machine to get as much information out of a 
sentence as a human would. 
A fourth class of problems concern memory. Any program 
must involve knowledge. H~mans do not use language in 
vacuo. Suppose I know that Sherlock Holmes is a tall, 
thin man. Suppose further that a fat, short man comes 
up to me and tells me he is Sherlock Holmes. If my 
memory and logic components are going full blast I 
immediately suggest to the gentleman that a)he is 
either lying, or b)could use a good psychiatrist, 
or c)he has a bad sense of humor. W~ would not 
like the computer to read a sarcastic sentence, such as 
"Surely they have a right to do unto others what they 
would not want others to do unto them" and file it away 
neatly. We need to give the computer a &ertain amount 
of linguistic sophistocation as far as irony, insinuation, 
27 
III-3 
and such go. This might seem overly optimistic, since 
most human beings lack this ability, but let me suggest 
that the goal of computational linguistics is to understand 
human capabilities, not reproduce them, something which 
can be done far cheaper by producing new human beings 
t~ru natural means than producing software in our labs. 
The only thing keeping us from programming ~omputers 
to~ for example, have a sense of human, is our peculiar 
delusion that we can't do it. 
So these are the problems that have not been the 
subject of serious research. Note that I do not mean 
by this that no one has ewr looked at them and 
found anything out. ~en Newton was Platonian enough 
to realize that nothing new is ever discovered under 
the sun. But no linguist operating in terms of a 
formalized or quasi-formalized system has studied these 
problems very much. This is not to say that certain 
conclusions about the future construction of a theory of 
language use cannot be drawn from ou~ present ignorance. 
T~e rest of this section will be devoted to how we 
with our Neanderthalic knowledge of language can outline 
a decent formal theory of 'la parole', something that 
we would want to do, I think, even had the computer 
never been invented. (~nd of sermon.) 
One question which arises ~ere is what the nature of 
underlying semantic structures is. Do people think in 
trees? McCawley in his article on the base rejected the 
notion of derivation. Instead he instituted a system of 
"node-sdmisslbility conditions". These are actually 
conditions on the well-formedness of trees. Any object 
meeting these requirements is a w ll-formed tree, other- 
wise it is not (although I have yet to settle in 
my own mind whether an ill-formed tree is still a tree, 
Just as I have been confused about whether an Ill-formed 
sentence of English is still a sentence of gngllsh at 
al~l.) Each NAC has the form 
<a; BC> 
which is read, "a node A is admissible if it immediate- 
ly and exclusively dominates a node labelled B and a 
node labelled C." NAC's generate trees directly, as opposed 
to rewriting rules which, in Choms~y's system, first 
go through a derivation, from which trees are then construct- 
ed. But the important point here is "Grammars are written 
by fools llke me, but only God can make a tree": meaning 
28 
iii-~ 
that linguists need not concern themselves with the origin 
of trees to discover their properties. 
Of course, if we are to be manipulating semantic struct- 
ures, we are going to have to be concerned with where 
trees come from. A more basic question is whether the 
kinds of trees generative semantics claims to be semantic 
are reasonable semantic structures, that is, whether the 
investigator in artificial intelligence, for example, 
could live with them. I think there is a very good 
chance that this is the case. The basic elements of these 
trees are as follows. We have referential indices referencing 
individuals. I think that in any system we will need a 
device such as this. Both these indices and larger entities 
called senetences or S,s can be dominated by the category 
N. I think again that any system will heed to consider 
sentences recursive in this way. Then we will need predicates 
of arbitrary "weight", tho' in natural language the number 
of N's associated with any predicate V will undoubtedly be 
rather small. One possible counter to this is obviated 
if we assure that we have ways of referring to sets. Then 
we can define S as a V and associated N's. This is not 
really a bad scheme. 
Where it does fall down is in its failure to reflect 
~yper-propositional relations. The conceptual universe 
of a person is not a bunch of unrelated trees or 
sentences (propositions). We will want ways to connect 
the Napoleon of "Napoleon ate cheese" with that of "NapoEeon 
hated Elba". Thus the conceptual universe is a network, 
with a far more complex structure than our underlying semantic 
trees. We therefore need some set of rules for isolating 
part of this network to serve as the underlying tree for 
some surface sentence or set of sentences, since it may turn 
out from our study of the structure of discourse that 
the unit of generation is larger than the sentence. 
More will be said on these matters in section IV. 
29 
IV-I 
PART IV. The understanding 
machine. 
One basic goal of research into computational linguist- 
ics might be to investigate how information is extracted 
from linguistic source data. (ultimately this ties into 
such questions as that of automated abstracting.) That 
component of our projected understanding machine which 
will model the information abstracting process let us 
dub the "info grabber~. 
The info grabber of course is not isolated. It will 
have to be connected with a logic component and a memory 
with which it will interact. 
Nor is this the whole picture. As shown below one needs 
also a way of encoding the semantic output of the_logic 
component for later output as linguistic data. Therefore 
the whole system will look like: 
I~i nguistlc\[~INFO "' ~LoGIc ource I1"~1 ~RAB- ata (LSD)~ BER ~output If SPEW- ~ata (LOD)II ~ER F 
Notice that I have dubbed that component ~ahich synthesizes 
the LOD the "infox spewer". 
W e can regard the above as a reasonable model not only 
of an understanding machine, but of the speaker. The above 
model would certainly be of use in the study of the use 
o f a natural language as a computer input-output language 
both for programming and for other applications, such as 
interaction between student and teaching machine in an 
educational program. I have made some study of such a sys- 
tem, which I called EASIOL (English as an Input-output 
Language), taking into account the results of the two 
studies I know of which approximated what I was after, 
namely Daniel Bobrow's STUDINT program, reported on in 
"A Question Answering System for High School Algebra 
Word Problems" Proc. FJCC 25, !9641 and in Scientific 
American S~ptember 198b, pp. 252-260, which was BobroW's 
30 
IV-2 
research for his doctorate. Bobrow modified LISPo_in the 
direction of COMIT, walling the hybrid METEOR. Re 
system he evolved has a fair amount of flexibdlity 
and generality, and can doal with many kinds 
of problems expressed in st~llzed language. I might 
criticize Bobrow for his naivete over natural language, 
but since I am even more naive about information 
processing I will not do so. 
A second system which I have heard later evolved into 
a more general system, is the BASEBALL program reported 
on by Green, Wolf, Carol 0homsky, and Laughery in the 
Feigenbaum-Feldman voltume, Computers and Thou~t. This 
system bases itself on a ra~-h-er stylized t~e of data 
structure. I have not followed the prc~ress of 
either of these projects, but both Betray inherent 
faults that made it unlikely t11at either could form 
the basis for a more general system operating on 
actual discourse. Nonetheless, these systems are very 
convincing for those Who think that language is the 
sacrosanct birthright of human Beings and that computers 
will never Be able to hahdle such tasks as writing abstracts 
of articles. 
The above model is also a reasonable model of human 
speakers (if we forget that people differ from machines 
in essential ways -- vive la dlff@rence.T ) 
The first part of the "info grab" is the read-ln. Hope- 
fully this will someday Be done by the machine itself, via 
optical reader or speech analyzer. I think that 
research on readers and a~alyzers has in general been 
unhappy because of a failure to realize how complex 
recognition b~ humans is. Recognition is not simply 
an ootlcal or auditory problem. All levels of language 
must interact in the process." it is well-known that real 
speech is more easily handled than ~pproxlmants 
to speech, x~this can only be d~e to the recognition 
process being cyclic and operating slm~ltaneously on 
all levels. The slmpllst recognition routines would 
involve something lik~ : 
$1 
IV-3 
SIGNAL ---~ ~oise Filter~ 
(LSO) -----=' ~ Segm~ter-- | 
I 
IMORPHOLOGICAL I ~I ~ 
i ANALYZER I ~ 2~ ~" 
Indeed, we have to connect up t~e logic and the memory 
to this system. Below is a real sample of my hand- 
writing when writing rapidly. 
No recognition routine, not even my own human one 
can at all times decipher this garbage. Redundancy is 
pretty near nil and such words as '!of", "as", "a", and 
"or" tend to be homologous. What a human reader can't 
do, we can hardly . expeet a machine to do. 
But humans can gmess from context what a word must be, 
and then see if the squiggle on the page is close 
enough. This involves both syntactic and semantic 
recognition, and if we ever want machine reading of 
handwriting, we must give the machine this capability. 
But suppose the reader still can'~ handle the 
writing? I suppose then we want ~t to get the logic 
component to intiate a question such as," What is that?" 
$2 
Iv-4 
That is, we want the computer to be able to go 
thru the whole set of levels. This will necessitate 
a much more complicated program than those around 
today, incorporating a greater amount of linguistic 
expertise, but undoubtedly it is necessary. 
Let us assume that the info grabber has grabbed the 
info, it ~_ll have (1) to store this information in the 
memory, and (2) ~et the logic component examine the 
information. Suppose I know that Richard Daley is the 
mayor of Chicago, and I read in a Chicago newspaper that 
the ~ayor of Chicago is the greatest man in the world. 
The LSD m~st somehow be so stored that I can retrieve 
from my memory the fact that RichardDa__~l~ is thot~t 
to be the greatest man in the w---~~t newspaper. 
This raises . the question of how to convert 
underlying semantic trees into subnets of the semantic 
network of which memory probably consists, hmny of the 
features incorporated into • Sidney Lamb's conceptu- 
al networks will, I think, be incorporatable into the moddl. 
In particular, all occurences of a particular entity (con- 
cept) will have to be linked in some way or identified. 
In a sense info grabbing starts by analyzing the LSD 
into semantic structures, and ends by synthesizing these 
structures and those already in memory into a new memory 
network. 
One point that should be made clear is that all informa- 
tion will have to be represented on the same level. That 
is, both the program and the data will reside in the same 
memory net, as in a computer. Reading an algorithm in 
a book, the machine will store this in its memory Just 
as it stores part of its own program, and it will be 
able to either quote the algorithm later as linguistic 
material as part of info retrieval, or use that algorithm 
as part of its own logical operations. There is some 
question the as to ~hether this quite ideal machine could 
actually function in this way. But human beings are like 
this in some ways, and it is part of their language capa- 
bility that they should or could. 
The process of info spewing is a reverse of the 
info grab. The logic component will initiate the 
spew, using part of the memory net and selecting 
one or more underlying trees to spew out. It will then 
3S 
IV-5 
go through the derivational process and ultimately genera@e 
an actual string of sentences. Perhaps feedback will 
enter here, so that the machine can utilize part of its 
own spewings as immediate LSD, although it is hard 
to see why the machine would need to do so, altho humans 
are constantly correcting themselves mid-sentence. 
An obvious question is what the role of generative 
semantics in all t~Is. I think the experience of CL 
has been ~ in general that ad hoc programs don't 
work. W e need a basic linguistic theory. I think 
generative semantics is the best bet. But as 
I noted, it is a theory of cometence. We will need 
to modify it. I think we need to 
l) admit rules of non-recoverable deletion, 
2) admit rules for hypersentential constructs, 
and 
3) build strong interactions with lo~ic and memory 
components. 
In particular, the relationship of underlying semantic 
structures to conceptual networks will have to he 
investigated in depth. 
If the hypotheses of the GS linguists are correct, then 
we have a simple but powerful basis for programs directly 
transforming language source materials into semantic 
information usuable by programs. For example, if 
the semantic structures turn out to be universal, they 
can servq as a pivot or intermediary for the currently 
out of fashion goal of MT. 
$4 
