Aaacriua Joarnd of Compntational Linguistics 
Mi crofi che 54 
MULTIPLE ENVIRONMENTS 
J4NUSZ STANISCAW BIEN 
Institute of Informatics 
University of Warsaw 
Palac Kultury i Nauki p. 837 
00-901 Warszawa Poland 
Copyright @ 1976 
Association for Computational Linguist ~CS 
The Daper describes a preliminarv stape of ~uthor'n innuGryv 
\\ 
Aimed at in-tegxating the 'Lpossible wo-ld. approa-.h with the 
idea of treating umerances as orog-ams. It is cl:?imed that 
providing ~1 onhisticated f aci litien fcr- mani pulating possible 
world'' descrintions should be one of the main concerns in 
designing a natural language understanding system. The logical 
\\ & 
notion of possible world" has a close counte-rpa~t in the 
computer science nntion of the environment of expression 
evaluation. The i?.ea of treating  utterance^ as programs 1s 
generalized bv allowing enui*onmt! ht s~aritohing during tha eval- 
uation of an utterance. A model of natural language, based on 
multiple envi-ronments in the sense just mentioned, is outlined 
in terms of computer science. A rough classification of envi- 
ronment tyoes is given. One sitructure of environments in dew 
voted to keepine track of the direct and indi-rect speech re- 
cursive quotations. Another structurG is asshgned to every 
person involved in a discourse or mentioned in it; it is used 
to handle belief-sentences, lies and promises. A third type of 
environment is used to represent the structure of topics in a 
discourse. Advantages of the advocated approach, called the 
rnultinle environment model of natural language\' are denon- 
strated in the discussion of well-known problems of ~f-rence 
ana presupposi-tions. 
1. Introduction 
2, DSscouyses aa programs 
2.1. Utterances as programs 
2.2, The notion of discourse 
3. Disc-burse processing 
2.4. Ambiguities 
3. lul tiple environments 
3. I. The no tion of environments 
3.2. Person environments 
3.4. Impression envir~nments 
3.4. Choosing an environment 
3.5. Topic environments 
4. Running an utterance 
4. I. Designators 
4.2. Pointera 
4.3. Presuppositions 
5. ConcZusions 
6, Referen~es 
The paper presents neveral ideas cn how Do describe natural lan- 
pages for a language understanding system. Some of them are 
similar to those advocated by Zlakoff (1968) and Mor~an (1969). 
They have been derived by the author independently (~ien 1975) 
while exp1ori.n~ the D~vies and Is~rd (1 972) ~PDTO~C~ of treating 
utterances as programs. 
The subject examined in the paper is i tself broad and encom- 
passes many controv~rsies;however, it is not the author s intent 
to make a case for the ideas presented. firstly, the limits of 
the paper do not permit a proper discussion of the pros and cons 
for each solution proposed; and furthermore most of these prob- 
lems have a tnadition dating as far back as the Itfiddle Ages, 
in some cases. And wxondly, the author has not yet developed 
hll of his own concepts fully enough to warrent a detailed 
presentation. Instead,the paner seeks to present the simplicity 
and generality of the proposed approach 
The wner is an enlarged and modified- version of a talk 
delivered at the Fourth International Joint Conference on 
Artificial Intelligence in Tbilisi. The modifications envolve 
mainly the terminology and the form of presentation; the only 
essential change of some importance is the different treatment 
of the first and second person pronouns. 
Most of the examples in the paper are direct quotations from 
the referenced literature; herein, some are employed somewhat 
differently than was their original intent. 
2. Discourses a_ proarams. 
2.1. Utterances an anoarams, 
1% is now obvious that the human ability to use language io re- 
lated Closely to intelligence itseif.Nevertheles3,the complexitv 
of hatural language is still rather underestimated by linguists, 
which results usually in using relafively primitive tools for a 
formal description of language. Although such formalinms like 
e. 6. transformation&l grammar may be theoretically adequate, 
from 
a practical point of view they are too cumbersome ; (i n my 
opinion writing a transformational grammar may be only compared 
with programming a sophisticated system exclusively in an as- 
sembly language). The main merit of krtificial Intelligence for 
the development of computational linguistics lies in suggesting 
a quite new way of thinking about language. It consists in 
shifting the at tention cTf re search from linguistic competence 
to linguistic perfomance and zreating the latter as an opera- 
tion of a real or imagined language ppocessor, which m turn 
can be discussed in terms of computer science.':'inoarad (1 972: 2) 
claims even that the best test of a complex model of natural 
language is to implement it as language understanding system. 
Although he is basically correct, in the present state of 
art, the objection posed by Charnialc is often vsl-iu. kelmiak 
(1~[2:2) noticed that most programs in Artificial intelligence 
handle only a few kindu" oT selected test exampleo. Decidin~ 
that a program can be extended in nome easily imaginable way to 
handle more extunple:: or more sophisticated cases regtxires prac- 
ticnlly the same procedure as verifying a non-programed theory. 
Therefore, I treat ~inograd's postulate an a long-krm aim, and 
at this moment Z advocate a less ambitious strategy: to use as 
much possible of the computer science intuitions in ??a tu2al 
Language description. This is in fact also the approach of 
Longuet-Higgins (1972) ~vho states that natural lang~age ut- 
t erances are just programs to be run in sur brains. 
Some interesting analogies between lahguage ~mderstsnding and 
rumsing a POP-2 program have been shown e. g. in (3avies, 
Inard 1972). I pursue this approach in another direction, 
charocterised by an intensive use of the notion of environment, 
In the earlier stage of the inquiry, represented by (Bien 1975)~ 
I thought that all the environment manipulations which were 
necessary for the feasibility of my ap-proach could be reali~ed 
bg rteans of the Bobrovi and 3egbreit rnl~ltipl envirollments 
primiSives (1 972) ; theref ore, I have introduced the term 
ti 
~~ultiple environments nodd of natural language , Now I am not 
sure of it, because I see reasons to use e. g, cross-world 
bindings , whose relation to the Bobrow-Vegbreit primitives is 
not yet clear to me. Anyway, I still use the term multiple en- 
vironments model of natural languages because it charac terises 
well my approach even if it is to be understood only metaphos- 
i cally. 
It should be noted that considering all utterances as a kind of 
imperative is not a new idea for linguists; it can be found 
9. 4. in (Wierzbicka 1969), but to my knowledge such lo eas had 
no practical impact on linguistic re searcn. 
2.2. The notion or alscourse. 
The notion of di~lcourse (called also coherent text) is a 
rather vague one. I will try to clarify my use of the term 
by discussing se veraJ. aspects of coherency. 
k'irst, there is a type of coherency which 1 shall call textual. 
It is realimd by these inter-sentence and inter-ph~ase lihks 
which are visible in the text surface as some lexical items or 
syntactical featmes. Surface realisations of these links I 
shall call pointers. A simple but very important class of 
pointers consists of Eronouns understood in a broad sense, 
including pro-adverbs et c. There are also pointers peculiar to 
given languages; e. g. after McCawley (1971) and Iaard (1974) 
it is reasonable for English to treat the Past tense as a 
pointer, because (isard 1974) it acts as a form of definite 
reference to a past situation on which the attention of the 
conversants has recently be= focussed The presu~posttions- 
often function ~imilerly to pointers, but I think that presup- 
positions are essentially different from  pointer^ and I prefer 
to account for them in another my, 
A second type of coherency I shall call s.ituetio~.al, The situ- 
ation of a conversation can influence the coherence of the 
message in two ways. First, it can supply values for these 
pointers which are not assigned by the text itself. It is the 
case of e. g. [Isard 1974a) 
(1) Be careful, he might bite you. 
said while the adressee is near a dangerous eninal. Such an 
utterance can be easily transformed into a textually coherent 
one by introducing a narrator. The second type of situational 
coherency is more subtle ,it consists of applying the addresse ' s 
knowledge to fill up some relations mi tted in the sender s 
message, This is needed e. g. in the text (Bellert 1972:79) 
(2) ~nn's eldest son has left Warsaw for a scholarship study 
in the Sorbonne, 
(3) France is an interesting co-mtr~ to study in. 
where the knowledge that the Sorbonne is a French universrty 
has to supply the missing link. In general, a text is situa- 
tionally coherent only relative to a given domain of knowledge. 
In practice we often communicate our ideas by means of 
non-coherent texts; the communication succeeds only beceuse 
the addressee modifies his beliefs for the purpose of making 
the text coherent relative to this updated domain of his 
ae~ief~, Because he does it only if he decides more or less 
arbitrarily, that the message has a meaning, such texts I shall 
call voli-bionaril.~ coherent,. h typical example is a text witl.~ 
3 sentence which carrie.~ brand new information by means of 
presuppositions. The existence of such sentences ha s been 
pointed out by Wierzbicka (1969), Bellert ( 1972:79), recently 
by Karttunen (1974:191) who gave the following examples: 
(4) 
1 would! like to introduce you to my wife. 
) We regret that children cannot accompany their parents 
to commencment exercises, 
where (4) presupposes the existence of the wife and (5) 
that its complement is tme but both sentences are used 
in situations which do not satisfy these p~ esuppositions. 
Tor the sake of completeness it is necessary to mention the 
situations, Where the text contain pointers without values, but 
they are considered by the addressee as not relevant to the 
matter at hand. This situation seems to happen only in literary 
texta. 
2*3. Discourse processing, 
I will discuss below the main levels of discourse analysis. 
I ignoie discourse generation for two reasons. First is a 
theoretical one: I Peel strongly that it is the analysis which 
bs the primary activity and that the generation is driven by 
the evaluation of the re-analysis of a generated text. Second 
reason is a practical one: at the present state of art we have 
much better insight into the analysis processen than into the 
synthesis problems. 
According to the present views I take for granted that the 
analysis consists of a set of cooperating processes performing 
different task, in pa~ticular the syntactic, semantic and 
pragmatic analysis, By a - level mentioned above I mean a se c U.L 
such processes which manipultite the notions of similar type, in 
the intuitive sense of the same degree of abotractness relative 
to the physical message. 
I distinguish four levels. 
The level responsible for extracting relevant information from 
acoustic signal or a visual image I call the sorbtion. I men- 
tion it here only for the sake of completeness as I have no- 
thing to say on this subject. 
The second I, 11 t::e only rsint uf interect of the 
present paper, I call the interpretation. I mean hy it the 
process taking as data the results of sorbtion (of course, it 
does not mean that the sorbtion is to be executed before the 
interpretation; the sorbtion should supply partial results on 
the request of the interpretation) and yielding some value in 
the formalism used in the system under consideration for the 
representation of knowledge. For simplicity 1 assume here that 
the knowledge is represented in labelled graphs stored in a 
classical way in a 1PLRNluER-like associative data base 
(Hewit 197 1 ), e.g. the value of the sentence (Charniak 1972:83) 
(6) Dill got the ball before he went to the park. 
may be something like 
(7) (Nl BEPORE DT2 N3) 
(NZ GET ljILLl BILL3) 
(lu3 GO aZLLl pARK1 ) 
It is often assumed that text processing by a language under- 
standing system consists only of those two levels or their 
gquivalence. I insist on the necessity of two additional levels. 
Yirst of them I call ,iud,gement.Thi; iz the lctvel 1.2 ,]Q:-. Tale Lo: 
keeping the beliefs of the aystem consistent. As long as triv- 
ial worlds are considered, this level can be integrated into 
some systematically performed uata base bookkeeping \'hen we 
start to model more complicated worlds, we will face the prob- 
lem of theoretical ur practical undecidability of bookkeeping 
problems and therefore this level is to be thoroughly control- 
led by the system supervisor. Such a solution agrees 1~Lth the 
t 1 
Intution of Marciszewski ( 1972: 180) that umst beliefs are 
spontaneous and that it is the entertaining of a belief with 
the awareness of non-accepting that requires a special act of 
giving up; the suspension of judgement is therefore an act more 
sophisticated than spontaneous aelief 'I 
The fourth lavel, which I call the internation, should be de- 
signed to memorize the facts marked by the jutigernent level as 
important enough to be   to red. As far aa 1 knolv, the investi- 
gation of the probleds related to this level has been neglected, 
the only exceptions being the works of Chafe, in particular 
(Chafe 1973). 
All the recent works on speech understandinr; as well as on 
discowse analysis show that the result of interpretation is as 
a rule ambiguous. For particular domains of discourse we often 
find some particular rules to disambiguate sentences, but the 
find solution consists, in my opinion, in formalizing and 
implementing general pragmatic rules, which I shall sketcn 
b e L OII~. 
The highest priority rule should be the rule- of c-ohe_rencs, it 
says that this interpretation of a discourse is better wi:ieh 
yields as the value the more dense graph.The density of a graph 
can be computed as e. g. the ratio of graph arrows to the number 
of nodes; if our graprls are frame structures in tr.e sense of 
Vinograd (1 974), we can compute the ratio of the mportant el- 
ements filled up to the important element slots left unas- 
signed. I feel it is just the rule which chocses properly the 
referent of the last sentence "she" in the examples (Charniak 
1972:56): 
(8) Penny wanted to go to %ill's party. 
~vlother had to Cell her that she had not been invlted. 
(9) When Penny heard about the custume ball $he started 
thinking about what Mother could weas. 
Idother had to tell her that she hhd not been invi bed. 
In general, the net effect of the col~erence rule vci.11 be thnt 
son-eticles the re Cerents of different nolzn l'.x3;hr:e.* (or'a in gen- 
~al, the lint,wCCal;ic *leans which I cell designdtoso) arc cox- 
lapsed to foxa one object Tor tht: sake of the higher dexlsi-ty of 
the result. Therefore the rule ~isil also help r to ilnl~dlc 
p,~e3txpponitions properly, ;le may treat every prcc:uppoastion as 
carrying brand nec. infor'nation and leave for the coherenay 
rule the task of collapsing eventually the presupposed facts 
with the facts already known by the system. 
Second rum I call the consistency_ rule. I mean by it the 
simple but important rule : if one interpretation of an ut terance 
ia inconsistent , look for another interpret ation, It explains 
why for the utterance (Russel 1905) : 
(10) I thought your yacht was larger than it is. 
one should not redct by saying 
(11) No, my yacht is not larger thm id; is; 
or why we treat the sentence (McCawley ?y67): 
{lr) Boris said that he didn't luss the giri who he Ussed. 
a8 the inf orma-tion thdt Boris lied ari'd not that he ut-teped a 
non-consisted utterance , 
The t3ird and moat subtle rule v~i~ic!~ ]sac the lo\*;: t primlt- 
\ 
C? i Tf 1' VtT, 
T: call the eft'icislwy rulc. . YCL:; 1: f icn J t,. 
kduine, can be found in A:dukiewicc () filis ivl~ '-tnSc;7 
$hat if we kave to choose between twc interprctntior.. cf R :eW*=- 
->; 
tecce, \:F c'l:z~~e the cther interpretatLo:~, 5.- 3:?b LU 
on the a~runlption thst tbic sv,..rlder wa:: a~:.\rose a' 413, :-;~U~QIV-I:.,~ 
p os::F1~i: i tic3 und zcnsi,~urly uzed t!:e r:slmt: 2 :;::-:icr? k tt.3 sar,:.~.:i~*~: 
J 7 \- 
t c trnnsmi t the newage es;lz*cc: L':le :; -2p ":iL: 5ezAt c--q% eL 
-- & 
- 
B 
- 
on the beao of 1;h~ ~ffj ~5 ~>XICY 1ule pref ~.r interpz.e-$r>- 
tion of ( I j) wi~ici~ is equivalent i o 
mw 
--.I 
(15) Caesar knew tnat -3orne lies on P ~;Ser md that -.me 
is the capital of the Fopes. 
Incidentally, in sore ,ci t71ations the ef f iciencv V rcl-e my suggest 
for the sentences similal to ( I?) zn inter~r.etaticn + analogical 
to (14). Let us assume for exsmpld thet Stanley is so1viz.g 2 
cros:q:vord puzzle and rlas to fill in a pattern spec;-Ee.3 5.: 
KI - \I 
.'+fir 
clue the river on vii?ich lies the capital ol the L ces ;~~~~~...Iby 
knov;s tk~t Eone is the capital of r3i)~ ~IJ$* $QPE. xct 
know that Rome lies on the Tiber. IIe may ask John about the 
name of the river and receive the proper answer. IIow, r:hen 
Ftanley is asked by somebody 
(1 6) Have you properly filled in this pattern? 
there is a quite natural answer 
(17) Yes, John said that the capital of the Popes lies 
on -the Tiber 
To swarize,I think the only solution to the ambiguity problem 
is the breadth first search in the sense of Charniak (1Y72:75) 
with the above-mentioned rules used to evaluate the inter- 
pretdtions. It does not mean that I neglect the need fo~ tl:e 
rules peculiar for particular discourse domsino, They are nec- 
essary for efficiency of the interpretation process and they 
should drive the search,but they may not be allowed to override 
any interpretation, as is the case in the ';!ilks preference 
grammar. For the sentence (Wilks 1974:32) 
(18) The hunter licised his gun all over, and the stock 
tasted e ape cially good, 
the real world knowledge would probably cause the interpreta- 
tion with the "soup" sense of stock to be found first,but the 
coherency rule would properly choose the interpretation where 
"stock'\ is understood as a part of the gun. 
Multiple environments. 
4 
2 1 , The notion of envirolment , 
Tn nctuL3l 1:mgunge understanding :~y:;temo the ftjc ts st ol*cd in 
the zystem memory are classified accordin(: to their ontolog~cal 
status in R very rough way. Usually they are split into the 
classes: past versus prcscnt nnd reality- versus po ssible 
future ; the only exceptions are the system for playing tic- 
tac-toe of Isard (1 974), Sosiii and 'Jeischedel ( 1973). In the 
linguistic literature one can easily find the ideas of pos- 
sible worlds used to handle the modal concepts,but more subtle 
possible woylds blassificatloas was discussed, to the best of 
my knowledge, only by Lakoff (1 968) and Morgan (1 969). hIy clam 
is that we need a very sophisticated classification schemae 
for the possible worlds features. 
A set of facts to which I assign the same ontological status 
I will call environment. The tern is borrowed from computer 
science, where it msans all variables access~ble from a given 
program point together with their values. My use of the term is 
justified by the fact that the access to the system memory is 
ussually performed by matching a pattern against an associative 
data base, resulting in binding the free variables of the pat- 
tern to some values found in the memory; access to different 
environments in the sense defined above nay result in different 
binding of the variables, which is also the case 7th the 
environments understood in tne computer science uenr e . 
Por the purpose of the present paper enviromuent!: can be 
thou~ilt of 3; L:ubsetr of a TLlKl!Ti3li-like data bsoe wi.t'n contenr; 
specified by some filters in goo1 statc ients. In reilit:,. such 
an implement~tion! would probably be inefficient and theref OL~ 
it i.9 necessary to look for another solut~on, ::tar tine; with the 
notion of co:ltext in Conniver ( Suosman, McUermot -19 7%) and 
QA4 (Rulifson et a1 1972). 
The crucial powt of tile presented approach con~iot:: in nL- 
lowing quite complicated access envirc:-merit ctx~uctures , to 
\\ 
stress this fact T use the tern "nultipl% environments . 
A sophisticated lenguagc understanding system has to take into 
account the fact that some other beings also have the ability to 
use na$ural language, to remember events, to make inferences 
etc, For the sake of simplicity any real or fictious being 
capable of using a natural language will be called a person. 
Practically, persons will be either humans, engaged in a dis- 
course with the system or humans mentioned in such a iiscourse, 
but we can mean by personE also another language understanding 
systems and robots, persofinications of an'imals from chilire- 
tales, etc. In the lack of arguments to the contrary, the oys- 
tern will assume that the language using abilities of other 
persons are identical with those of itoelf. Therefore it can 
easily simulate relevant aspect of other persono mental 
processes by running recllrsivelg its own language proceaa&ng 
routine in a suitable environment. For example, if the system 
perceives the ~entence (13) said by Pred-tq Stanley: 
(19) I like YOUT boo:i. 
if -\\ \\ I\ 
.L and you are respectively bound to Fred apd \Stanley; such 
treatment allows it to obtain the correct value of tile sen- 
tence, which may be represented as: 
(20) Fred likes stanley's book. 
In the text: 
(21. r) Frank said: 
(21.2) " Ben I came back, John was already waitlng for me 
and asked: 
(21.3) 'HOW are 
the clause 1. I) may be evaluated in the snvirbnment of the 
system itself. The mention of Frank causes the environnent of 
\\ \\ 
Frank to be created or recovered with the prounon I preset 
to Frank ; the interpretation or ( 21 .2) is done by the recursive 
call to the language processing routine in the environment of 
Prank. The mention of John yields the environment of ~ohn,which 
is embedded in the environment of Frank; therefore from the 
point of view of the system, it is the environment of John as 
described by Frank. This person environment has the pronoun 
I" preset ao usual to the respective person and the pronoun 
you\' is set to nis current interlocutor, i. e. PxM~~, 'l'he sen- 
tence (21.3) and other sentences reported by Frank as said by 
John are interpreted in th~s environment. 
;'lultiple pernon environmentu allow the oy~tem to keep track of 
3ifferences in person's knowledge, which has importance for 
neny reasons. For example, it is difficult to account for the 
use of the indefinite noun phrase in (22.5) in the text 
(~harniak 1~72:185): 
(22. I) Jack and Bill were outside flying R kite. 
(22.2) A strong wind came by and the string broke. 
(22.3) ~anet and   lice W~TC ''~ut~ici+? the house. 
(22.4) Janet loaked up and said: 
(22.5) Look Alice, there is a kite i ~?cag. 
1:-ithcut qarating the knowledge of the ~arrator and the ad- 
dressee from the howledge of Janet, 
By keeping track of the knowledge and oeliefs for every persor. 
in the discourse, we are able LO give an obvious explanation 
for the difference between factual and non-factual sentences, 
e. g. (Kiparsky 1971 :345): 
(23) John regrets that it is raining. 
(24) John thinks ttst it is raining. 
In the first case it is the environment of the addressee ?:rk:ich 
\i i\ 
is affected oy an evaluation of the phrase it is rainir~g . 
According to our treatment of presupposition,  he value of that 
phrase is added to the environment of tho oddreooec,  ere 
eventually mag be collapsed with another mention 31' tile rain 
An analogical process is independently performod in t:ie pnvi- 
Ponment of John. n the second case it is exclucively tile i?:-- 
vironment of John which is affectrtd by the evaluation of tho 
I\ 't 
phra:;e it is raining , while tile addre::-ee ' 3 environmer?t 
r emainx u~ichanged . .lo has been pointed out by Ido~gan (1"\.3) 
it is not true that non-factual eentencea have no presuppo- 
sitions. It can be easily seen in the following text: 
(25.1) John thought the door is open. 
(25.2) He thought he should open it 
which is inconsistent in the same way as the text 
(2b.l) The door was open. 
(26.2) John went to open it. 
The difference between (25) and (26) consists in the environ- 
ment to which the presupposition refers: the real world or the 
mind of John, 
It should be noted thdt the class of non-factual sentences and 
phrase is much larger than is usually assumed. For exemple, in 
the text (~harniak 1972:191) 
(27.1) Jack and Bill were outside flying a Ute. 
(27.2) A strong wind came by and the string broke. 
(27.3) Janet and Alice were outside the house. 
(27.4) When Janet looked up she saw a kite. 
a kite of (27.1) and (27.4) in reality refers to the same ob- 
ject, but in (27.4) it io referred again by means of the indef- 
initc noun phrace to mark Janet's ignorance about it, h'e lnriy 
eaoily account for it by evaluating thio noun phrase only in 
tho environment ol, Janet, 
Some canes involving non-Tactual interpretation9 of phrrine:: 
and nentenccs have been diocusacd by logiciap; nnd philo::o- 
phers under thc name of intcllkionnl (opellcd cvith 't' ) o~ 
& 
rntonoional (spelled with ) verbs, e. g. (IvIontaguc lcjrij3) : 
(28) John lookc for a unicorn, 
intensional sentences, e. g. (Ajdukiewicz 1959): 
(29) Caesar knew that the capital of the liepublic lies 
on the Tiber, 
*" , 
or belief-sentences, 3. g. (Tal~tec Id*- ): 
(33) ?~a be1i.c:~: s that you an I are sister::. 
A11 theae cases can be easily handled by meano of multlple 
~nvironment s . 
3Jrn~ression environments. 
h the above discussion of persons we have assumed that exactly 
one environment is assigned to every person. There are several 
arguments for splitting this environment into at least three 
mvironment s, which I call the environments of behaviour, 
pretense and knowledge. 
Distinguishing the pretense environment from the knowledge one 
is necegsary to handle the cases of lying, e. g. 
(3 I ) Fred is lying when he says he lilces Stanleyc u book. 
we interprete as meaning 
('32) Fred likes Stanley's book. 
in the Fred's pretense and as 
(33) Pred does not like Stanley's book. 
in the ~red'o belief, 
This distinction allows us aloo to handle the performatives 
along the lines of Lsard (1974a), i. e. the sentence 
(34) I bet you 2.5 p that it will rain tomorrorv. 
1s taken to be true, because uttering it changes the respective 
t 
pretense (which is equivaleat ts t;l2 rl~t;oc af pozl;. 01 rere;- 
ence used by Isard) and the judgement during the comparison of 
- 
the value of (34) with the content of respective person envi- 
ronment founds them compatible. 
The distinction between behaviour and pretense is more dis- 
cutcible, as it yields subtlety not needed in most applicatio~is, 
It is useful to handle e. g. slips of tongue and to distinguish 
* 
different aspects of what is saidU (Ziff 1972). 
To accout for such linguistic facts like e. g. the accept- 
ability of 
(35) John wants to catch a fish and he intends to eat it. 
and the non-acceptability of 
(36) John intends to catch a fish and he wants to eat it. 
it is necessarg,following Lakoff (68:7) to introduce additional 
environments,namely the environment of desires and the environ- 
ment of in"cel?tiol~s, 
t\ b\ 
The word knowledge" in thc term knowledge environment'' is not 
to be understood literally. In fact it describes the knov~ledee 
attributed to oomebody on the evidcnce of its behaviour, ntate- 
menta etc. This juctifies using the term impression environ- 
ment s to cover all the environments assigned to a per::on. 
Choosina an environment, 
From the preceeding paragrapn it follows that during discourse 
analysu we have usually severtil environments at hand. Xvery 
phrase is to be interpreted in at least one environment. This 
poses the problem of choosing the proper environments for 
the interpretation of a given phrase. 'Je will see that there 
exir~t several strategies to handle the probles, vihich differ 
in the amount of computationttl resources (tine and memory) uoed 
this fact may be considered to be the analogue of the hunan 
ability to change the amount of attention devoted to discourse 
undre st anding. 
Let us imagine a highly sophisticdted computer-aided instruc- 
tion system. During a teaching session with a student the sys- 
tem models its interlocutor by means of a person environment, 
composed of the behaviour environment (used e. to "On!lr.? 
the student response time), the pretense and the knowledge 
environments, The knowledge environment is preset to Gome 
general knowledge and it is oystematically updated by t;,e ma- 
lysi~ of student uttenance~. If ke do not exclude the yes- 
sibility that the student lies, -e should interprete theill 
directly in llin knowledge environment. In any co-c , the pretense 
is useful to st~re the result:: of such perfornativcs 3:: 
otudent':: definitions of nyl:lbols etc. If the student i:uotes 8 
manual or a lecturer, it is necessary to creetc a new person 
environment and to evaluate the whole quoted passage in it '.Tr?en 
the system is going to say somethiiig to t:le student, it 21:suld 
,- -2 J ,- -, 
verify ~:li~ thcr the u"im;.o::z -- A - LA - *A - - A- -- Am .-,--> +-- 1111, 
m 
c~tudent. 1 verification is made by running its utterance in 
the student's environment, which allows it to compare the real 
intention o$ its utterance with the intention probably assigned 
to it by student. To be more strict, in the student's environ- 
ment the environment of the system as imagined by the student 
should be created, and the utterance run in the pretense of it; 
the results should be compared with the results of running the 
utterances in the proper pretense of the system. 
O~viously,this is a complicated and resourc~ consuming process, 
difficult to carry on with reasonable efficiency. 1Jewertheless, 
I think that processes performed in the mind of a talented 
teacher working at full capabilities must be similar. 
Let us now take ail example from the other extreme, when the 
system works according to the reliability rule , It may be the 
syotem from ow first example during a session of knowledge 
acquisition when discussing a subject with a teacher. Now 1% 
is quite posaible to introduce all the information directly to 
3 * 
the lcnowledge environment of the ,;yntcm, 1; I. L :.hc ti::ol s 
a credously accepting everything v~lirlt io said. It ir evidently 
quite easy both for humans and computers; in practice hunan.1 
apply this strategy mainly when they are forced to devote rnor3.t- 
of their computational power to other tasks of a higher 
priority. 
Between these two extrenes there exist many mixed strategies, 
where the system for every person of the aiscourse makes an 
independent decision, whether or not to create a new environ- 
ment, based both on the reliability of the person and the 
avallabilits of resources, 
r T-opic environments, 
The usual way of handling the reference problem is to compute 
for every noun phrase or pronoun a separate list of possible 
referents afid to use some heuristics to choose one element from 
the list. Another approach, advocated here, consists in storing 
all possible referents permanently available into a special 
topic environments. The items stored in the topic environments 
may be the symbols of physical object as we21 as some relations 
and other semantic data structures, put there during the eval- 
uation of respective,linguiotic constructs. In particular, thr 
indefi-nite dttscriptioi1~ of the type (3ellcrt 197':: 32) 
(37) It is a foreigner who is delivering espeech now. 
(38) One young boy has flunked his mtriculetion exem, 
are interpreted as declarations, used to create appropriate 
data structure3 to be put into the topic; therefore do r.ct 
need the Ref operator of Bellcrt (1372:3 ) introduced to ksndlr 
such examples. 
In every topic environment the items are rdered in ooxe way; 
the access routine 1 cn re quest the mbseqirent elements 
uf the topic in the very order; if the environment is exhausted 
and some new candidate for a referent is nee6ed, tae ac em 
routine switches to the respective super-envirorment. 
hkery succesful access to the topic cauees it-s pe~mutation,~vhich 
results 5-11 making the accessed item thc first element of tile 
environment. 
The essential difference bet~~*ec:: the traditional approach and 
that presented he17e lies in the possibility o"  st^. ct~rine, a 
topic enviror~?.ent. 
For example (~sard 1974a), in the text: 
(~9.1) ;'!hat did John say about; Dick? 
(39.2) He said that 
(39.3) he looked like a druaken giraffe cn ibe skates. 
there is co doubt that hhe" of i.39 '3) refers to Dick.-ile account 
ioz* it in the following way: every first use of a vero of the 
'\verbLm diccndi'' f-yp.l,e c3uaea a 1 topic environment to be 
created; tile environment is ntali, in our case to Dick 
and tlle~~ itcelf bec~mes an item of the current topic. Tr~e next 
mention of that act of communicatio~ cQU:ie:; the se~rch in tire 
current topi0 for the environment previously stored ; vhen it ic 
found, the reported clause is interpreted within it. Tn our 
ceoe the first word oi: the reported clause is -\;he pronoun "he\' 
which calls the access routine for a posoiblc referent. 4s tho 
first and only possible relerent of the current toplc is 1)i.c~ 
the pronoun is properly bound. 
In the caze of the text 
(40.1 ) '#Rat bid John sag about nicp? 
(40.2) iie sald that 
(40.3) te doubted whethlr irick would like it. 
tile pronoun will be also at first bound to Dick from the topic 
\\ 
But when the proper name ~ick" will be evaluated, an attempt 
will be made to shift ilick to the first positiou in the topic 
where it already is located. This is a violation of the elfi~ 
ciency rde, and this interpretation will e rejected. Rext 
possible refrent will then be obtaln from the super-environment 
yielding the correct result. 
There are some exceptions to the efficiency rule; in particular 
redundancy is desirable immediately after topic switching: 
(4 1.1) Ylhat did John say about Dick? 
(4 t .r) He naid that 
(41l.3) Dick Tooked like b drunken gird~fc or. i~e l:ates. 
Another al-gur~ellt for topic : t:*uctullint; is ti:c exist P:.~c iYT xi': 
(42) Jnnet, M.11 and 1 sis t ex1 tir.lc:l :.:ere cut ::l<e, 
it 
Janet Baid: 1' calibt keep this kitten. 'Jould you like to 
'I\ \\ \\ \ .,* 
hdve itr liclen? . Yes , mid Helen. Rill said A don't 
/<now, Remembpr ! t ~k Jcc ted to tphn t robin. >he 
" \\ 
would not ict us keep it . But L.other &aid it is cct 
\' 
\ i 
good to keep s robin indoors, sa~d flelen, It is ~ot 
I\ \ \ \ \ 
fair to the robin , Look, sald Janet, D, you want it 
or not ?\'. 
At this stage the topic problem is the least elaborated part of 
my multiple ewironments approach;therefore I am not a~le to go 
into debails here, Fome additional argument in favour of the 
topic environment ideas will be fcund in the paragraph 4.2. 
4, Runninrg an utterance . 
By desimators I understand the linguistic means used to refer 
to particular objects. Designators may be classified roughly 
into three classes: proper names; common names and description9. 
All three kinds of designators have been discussed for centuries 
by philosophers and logicians; our pragmatic approaah to the 
designators is based mainly on the works of ICripke (1 972) and 
Donnellan (1971). 
The characteristic festure of proper names is that their use 
must be preceded by the act of fixing thdreference. Contrary 
to popular opinion, in practice proper names we much more 
ambiguous than common names , because pr ORT names re f er to 
individuals and common namels to classes of  individual^, For 
example, there are many men called John and if weare 
to understand e. g. 
(43) John has come. 
we need to have the referent of John already fixed. Different 
types of object have proper names of different stability. 
Countries, towns, mountains et c. have of ten unique names which 
are rarely changed.Most human full names are also rather stable 
from the practical poiat of view.Forenames are so ambiguous that 
their referents have to be reset again and again in every dis- 
course, e.g. by quoting the full hame of the person in question. 
Such proper names as first and second pereon pronouns may 
alternate their meaning even during one discourse. To sumarize, 
proper names in fact name some recognition routine supplied in 
the act of fixing the reference. 
Common names do not refer directly to individuals, but they are 
names of characteristic functions of the respective sets of 
individua1s;their mealling is so stable that they can be assumed 
to belong to the knowledge of the world; and therefore, fixing 
the reference, except in some peculiar cases, need not be 
performed. 
Descriptions are compound names constructed nd iloc by npccial 
linguintic means, like qualifyirlg 3 common noun by an ad jcctive 
etc. They name 31no respective compound chnmctcri~tlc 
functions, 
All the desic;n;nators are usually used in tne referential way. 
That means that tile respctive procedure is cvnlunttJ in a 
proper environment to yield the intended referent. Such treot- 
meat agrees in particular with Donnellan's intuition that a 
definite description does not in itself refer to anything but 
only its use points to a referent. Therefore, c.e nay explain 
easily why the sentence (~onnelltln 1971:llO): 
(44) Her husband is kind to her. 
can be sometimes properly understood even if the man refered to 
is not the husband of the woman; such a sentence is just to be 
evaluated in the environment of false belie of the sender. 
Quite often the sender uses undersp'ecified descriptions, i. e. 
the descriptions which refer to a much larger set of individ- 
udls than it is intended, e. g. in Gnarniak ( 1972:72) : 
(45.1) Mother made some cookies and left one on a plate. 
(45.2) She put the plate on the kitchen table and went. into 
living room. 
"the pla~e'' of (45.2) by itself refers to every plate of the 
world. 
The addressee has in such situations to restrict the respective 
characteristic function to the object on his current topic; 
if it does not suffice, he need to eliminat8 the remaining 
ambiguities in the usual way, i. e. by meano of the coherency, 
consistency and efficiency rules. 
All designators except pronouns can also be use tn an attribut- 
ive way. In this case 
- 
they mean just their characterintic 
functions. For example 
(46) hiount Everst is Chomolungna. 
means that in the sender'a beliefs the characteristic function 
of fv'lowlt Everest has as well all relevant seatures of Chomol- 
ungma. In some other usage the respective characteristic 
function is to be adapted by the addreL,;ac to he JLL:*-X 11-5 
0 A' 
01 bale 3 <,it' L \--:- A%. 5 9 Ij t--7 
(47) Ile is a little Napoleon. 
More common is attributive use of common names and descriptions, 
b 
d. g. 
(48) The smith's murdered is insane. 
in the sense of 
(49) Whoever has murdered Smith, he is insane. 
Attributive use of indefinite descriptions is a1 so known as 
L\ 
nonspecific setting , e. g. (Charniak 1972:178) a kitten" 
in (50. 1: 
(50. I) Jack wanted a kitten. 
(50.2) Bill had a kitten and Jack offered to trade his ball 
for the kitten. 
(50.3) Bill wanted to keep his kitten, so Jack went to look 
for George who also had a kitten. 
(50.4) George was willing to trade so Jack got his kitten. 
It is intereating that the SI1RDLU program (Y/iinograd~l~7 2) 
treat indefinite descriptions,which are used only at t x~ibutivel~, 
'3. g* 
(51) piclc up a big block 
(52) find a block which is taller than the one you are 
holding and put it into the box 
just along the lines given above, i. e. as programs with a free 
variable (Xinograd 1972 : 1 30 1 , a1 though Pinograd himself de- 
scribes it, probably influenced by the logical tradition, by 
means of the variable bound by the existential nuantif~er 
(1972: 126). 
The characteristic feature of proper and cormon nouns lo thet 
they ore always to be run in a single environment. It is not 
the case with descriptions which mag require suitable ~plit- 
ting between several environments. It has been noted by 
Winograd (1972: 147) with respct to the tine reference, e. g. : 
(53) Idany rich men made their fortunes during the depression, 
(54) Idany rich men lost their fortunes during the depression, 
(55) Many rich men worked in restaurants during the depres- 
sion, 
\ 
In these sentences the rich men" phrase is to be evaluated in 
the present time environment for (53), in the past tine envi- 
ronment for (54), and the sentence (55) is ambiguou:: when 
taken out of a larger context. 
There are other ainilar case:q, e, g. the acctendc 
(56) Smith Knows thdt the friend of i~owal-hic s brother i:: 
is subject to several interpretation:;; acme of then art3 
(57) Smith kncwu Kov~al ski, ICowaZcki s bx*otL-:c~ r'In i,i~ 
.Criend of ~owalski's brother and Llnimlth lanows that the 
friend of I(owa1ski ' s brother is a vrrite13. 
(50) Smith does not know Kowalski, but he knows ~owalnki's 
brother and the friend of icowalski's brothel. md Smith 
knows that the friend of ~owalski's brother is a ivriter. 
Obviously, ' the interpretations (57) and (58) differ in the way 
\\ 
the phrase "the friend of Powalskf's brother is split be tween 
the environments of the knowledge of the sender and the ih?ow- 
ledge of Smith. 
Mnce we allow designators to switA environm~nts, we have no 
problem with so called nouns vvi tll erlpty denotdtion ; they are 
to be evaluated in the respective fictitious worlds, e. g. 
(59) I met him in the park by the sculpturt. of a faun. 
4f2. Pointers. 
In this paragraph we shall dizcuss the reference problem for 
the third person pronouns,the most important class of pomters. 
\\ 
We shall f'ormulatc a rule intendcd to substit~te the connand 
\\ 
rule xi' the tmnsf ox?i!ational grammar. The arguaento in J avour 
of our rule are oinplicity, bcttt~~ odequ~cy and ufiage of t21c 
demo11 facility whicn should be prerent in the sgstel:: for other 
reanons ( ~harniak 1972). 
First, we ahall remind that every accesr to the tcpic pe-nutc:: 
it by advnncicg the ac3crseed element to the very beg inn in^; 
of the topic; succesful evtlluation of e designator introducing 
~1 new aob ject puts the rcpre::ent~tL~:: 3f the obj ect also at 
the very beginning of the topic. Our rule stdtes that the main 
L 
c lause pronouns imrnediat ely execute access .o the !o;Lc, 
while the evaluation of subordinate clause pronouns rcay be 
suspnded if tllere io l~c suitable value for them ln the 
I? 
topic; all the pronouns s'noYbxld be resolved be t.Tbr ri:t. il!c c~f 
the main clause processing. 
Let us see how the rule works for examples from (:.lcCawiey 
1~/1:226). 
(60) 4fter John left his apartment, he we~t to the pool hall. 
The proper name " ~ohn" advances the J'orm's representation at 
\\ 
the beginni~g of the topic. The pronoun his" is evaluated 
because there is at least one suitable value in Xie topic, 
i.e. John ; if bounding of\'his" to John is rejected by some 
r\ \\ 
pragmatic rule, another possible referent is found, ne is 
set to it and the referent is advanced at the beginning 
of the t @pic, getting ahead of John , Vhen "her is evaluated, 
'\ 
it is us~elly bound to the same referent as his", which may 
be John or some other person. 
(bl) After he left his apartment, John went to the pool 
hall, 
" ~fter" introduces a subordinate clause, whii.h may be auo- 
\\ 
pended if there iz no quitablc value for '\he\' or "i7is in 
the topic; if the clause io suspended, it can be roomed :.dlen 
\\ \\ 
evaluation of Joh~ puts its valuc to the topic. 
(62) Joha -:ient to the pool room uf ter he left :.;is cpartnent. 
\\ \\ 
There is no problem with the sentence ; John sets the topic 
\i 
supplying a possible value for "he\\ and his". 
(63) He went to the pool hall after John left his apartment. 
The "he" is evaluated immediately; therefore the evaluation of 
" ~ohn" cannot influeace its value. The strong feeling that the 
value of "he" should be different from John is explained as 
the violation of the efficiency rule: if you can refer to an 
object by a pointer and there is no possibility of misunder- 
stqnding, do not refer to it by a designator. 
(64) A boy who saw her kissed a girl who knew him. 
I am anxious to see a sentence of this type in an authentic 
English text, not as an example of areference problem, because 
I do not see any circumstances under which such a sentence 
can be uttered. Bone the less, we can handle thehexample easily. 
First, 
a boy" is evaluated, yielding a per son environment v~hich 
I L 
is put into the topic. Next, who saw her" is evaluated except 
for the \her\' because of the lack of a suitable value for ;t In 
the topic ; therefore , the clause is xuopendod, '!!hen the rest 
of the main clause 1s evaluated and the peraon environm~nt for 
\\ 
H is created, resulting in an updating of the topic. 
Now the suspended clause can be reoumed and evaluated in paral- 
\\ 
lel with ti12 clause who knew hid' ; all the pronoun:: will be 
properly bound. 
Althougll it should be obvious, it is better to state explicitg 
that the value of his in (60)-(s3) is finally fixed by tile 
coherency rule. 
Analogically as with subordinate clauses, we can treat paren- 
thethical clauses and phrases, e. g.: 
(b5) In ~ohn's apartment, he smskes pot. 
(66) In his apartment, John smolces pot, 
and obtain the desired results, 
It is hoped that the above discussion together with works of 
Isard, Davies and Longuet-Higgins has shown the advantage of 
treaking discourses and utterances as programs. Below we give 
an additional argument in favour of this approach, based on an 
article by Karttunen (1974). He presented difficulties related 
to finding the presupposition for a compound sentence and 
suggested they b'e avoided by adopting the more dynamic approach 
of a recursive verification of the satisfaction of presupposi- 
t ion" condition. This idea can be ea~ily integrnted in our 
model, as it is test illuatratcd from examples 
(67) If Dean told the tru-tih, Nixon ia guilty too. 
(bR) if Haldeman is guilty, Nixon is guilty too. 
(69) If Miss Woods destroyed the missing tapes, Nixon is 
guilty too. 
The consequent clause in 911 of these prenupposca the guilt of 
someone else, but the presupposition of the irholc sentences 
J".? 
-?; 1 - ; , , $1,1,)- 
differ: (b8) definitely do-:: not prc;;~1i);)3.. .IL c,LLL- 
olio i- Ise vfkila (67) and (69) may presuppose it or not, depending 
upon the circumstances of their use. In our model the antecedent 
clauses are run before the consequent clause; therefore in (68) 
the presupposition of the consequent clause is satisfied by the 
antecedent and in (6'1) and (b9) it can also be the case, de- 
pending upon the other knowledge available. in the environment 
of the evaluation. 
, Conclusions. 
&en if some of the presented ideas may becoae obsolete throught 
further research, the usefulnew of the soph~ s*~ated environment 
B trmctures for natural language descriptions se ems evident. The 
notion of environment deserves a place as a crucial notion in 
a fully adequate theory of natural language. It is important 
that the environment structures postulated here involve quite 
complicated embedding of environments without any static 
restrictions on the depth of the embedding. At the preoent 
state of art, it is convenient to deocribe such struct6reo in 
. . 
terms of computer science, because e. g, t'o1~1e.l 1ot;i.c LA i.~ - rk 
only u 1 -~213. ': 1~1,s bc tneen di fi'erent po~ciblr coslds . 
This fact, together with some other advantages shorvn in the 
papey, makes a strong argument in favour of treating discourses 
and ut t erances as pr ograrns . 

References

(A Sukiev~icz 195b). Kazimierz Ajdukiewicz. Okres vrarunkolRty E 
inlplikacja materialna.Studia Ilogica IV, pp 117- 153,Reprinttd 
in (Ajdukiewicz 1965), pp 248-265. 

(~jdukiewicz 1965). ICazimz iZjdulriewicz. Jcz~k i poznanie, 
t ,XI, 7;Varszawa: P1!m, 

(4jdukiewicz 1967). Kazimi\erz Ajdukiewicz. Intensional Zkpres- 
sions (a lecture delivered in 1959, published posthumously) 
Studia Logica SS, pp 63-86. 

(Bellert 1972). Irena Bellert. On -che logico-semantic structure 
of utterances . VJroc2aw--Wars zawa-Krakoiv-Gdansk: Cssolineum, 

(Bennet 1975). Michael Ruisdael Bennet. Some extensrons of a 
Monttlgue fragment of English. Reproduced by the Indiana 
University Linguistic Club. 

(Bien 1975). Janusz Stanislaw Bien. Toward a Multipl Lhviron- 
ments Model of Natural Language. Advance Papers of the Fourth 
International Joint Conference on Artificial intelligence, 
Tbilisi, Georgia, USSR, pp 373-382 

(Bobrow, Wegbreit 1973). Daniel GoBobrow, Ben Wegbreit. A Model 
alld Stack Implementation of Multiple Environments. Comnunica- 
tion of the ACM , Vol. 16, No. 10, pp 591-603. 

(Chafe 1973). Wallace L. Chafe. Language and Memory, Language 
Vol. 49, No. 2, pp 261-281. 

( Charniak 1972) . Lhgene Charnialr. Toward a Model of children's 
Story Comprehension. Massachussets Institute of Technology 
AI-TR-226 

(Daviss, Tsard 1972). J. M. Davies, S. D. Isard. Utterances as 
programs. In (Meltzer, Michie 1972), pp 325-340. 

(Davidson, Harman 1972). Donald Davidsm, Gilbert Harman (eds.). 
Semantic of Natural Langudge. D. Reidel. 

(Donnellan 1971). Keith Donnellan. Reference and definite des- 
criptions. In (Steinber, Jakobovits 1971), pp IOU-, 14. 

(~eigl, Sellars 1949). Herbert Peigl, lfid Sellars (edo. ) 
Reading in Philo~ophical Analysis. New York. 

(Fillmore, Langendoen 19 71 ) . Charles J. Fillmore, D. Terence 
Langendon (eds.). Studies in Iinwistics Semantics. Holt, 
Rinehart add Wins ton, 

(Kintika et al. 1973). K. J. J. Zintika, J: N. . ldoravcr;ik, 
P. Supps (ads.). Approaches to Batural Language. D. Reidel. 

(~ewitt 1971). Carl IIewitt.lleccription and theoretical ancl;~-:--a:: 

(uaind r ch~.r.lntac ) oi 1 Tl:\XmTN!F1&: language for proving 
theorems and manipulating models in w robot. h<assachuoseto 
Institu~e of Technology !'\I-TR-d58. 

( Isard 1974) . S. D, Iflard, Yffiat twould you have done if . . , 3 
Theoretical Linguistics Vol. I No. j, 

( Isard 1 Q74a). Stephen D. 1r:ard. Zhanging the Context. Ag~~me- 
ographed. 

( Joshi, 'i/eischedel 1973). 1. ~r . Joshi, I. 7 1, ;:'elscheclel. S om e 
i~ri.11~ fo~ ";w TRoc"cel Tic-Tac-3s I.>- LC c~1~td ~UCYC: 
St3~1t;~t i L' ;; >L' I'redica-be Cornp3rneliC 3on~-+~~-1--'i~-- ri -%\-. L VALe Ldvaaz e 
Paparo of the Third Tnternational Joint Gunf erenc e zn 
tztiTicin1 ;r,telligence, Stanfol*d, US.?,, pp. 352-'r '' 

 arttu tun en 1974) .~a~wi I:a~~tt~cn.Presu~~ositions and Lingustic 
Context. Theoretical Linguistics Vol, 1 No. I/?, pp IS?-194. 

(~iparsky, Xiparsky) . Paul Iiiparsky , <jar01 Kiparsky . Fact, 
In (Steinberg, Jakobovits 1~71), pp 343-369. 

(iwipke 1972). Saul A. Kripke, Meaning and Necessity. In 
(Davidson, Harman 1972), pp 253-355. 

(Lakoff 1968). George Lakoff. Counterparts, or the problem of 
~eference in transformational grammar. Eeproduced by the 
Indiana University I inguistic Club. 

(Longuet-Higglns 1972). H. Christipher Longuet-Ifiggins. The 
algorithmic descriptionof natural language. Proceedings 
of the Royal Society London a. 18r, pp 235-216. 

(I.lercl~::ew~~i 197~). 'Uitold Marciazewsiri. 130d:jtawy logiczne j 
tcorii p2zckor~an. ,inr::zavra: T1!lN. 

(Kc Caviley 1911). Jnmecl D. Mc Cawley. Ten::e and 'l'irne Tieferonce 
in kgliah. In  illno no re, I,sngcndon 1~71). 

(Mc Cawley 1971). Jmeo D, l~!c Cawley. Where do noun yllroaes 
come from? In (Steinbert:, Jakobovits 1971 ) , pp 217-231. 

(Idc Dermott, Sus~rn~n 1972). Drew V, Mc IPer~~lott, 
Sus~man, The Conniver Referenre u bT~issachusae tc: 
Icstitute of Technology A1 Memo c59. 

(Meltzer, t~llchie 1972). Bernard Nleltzer, Donald X'i:ichi.e (eds. ) : 
Machine Intelligence 7. Edi~burgh: University Press. 

(Montague 1973). Richard Montague. The Proper Treatment of 
Quantification in Ordinary English. In (Hintikka et al. 
19T3), pp 221-242. 

(Morgan 1369). Jerry L. Morgan. On the Treatment of I'resup- 
positions in Trensformati~nal Grammar. Papers from the Fifth 
Regional Ttlrleeting Chicago Linguistic Society. 

(Rulifson at ale 1973). J. F. Rulifso:~ "t al. QA4: h Procedural 
Calculuo for Zntuitive Reasoning. SRI dI Center Tec'mical 
Rote 73. 

(Russel 1949). Bertrand Russel. On Lenoting. Reprinted in 
(~eigl, Sellars 19 49). 

(Steinberg, Jakobovits 1971). Danny D. Steinberg, Leon tl. 
Jakob~vits (eds.). Semantics, An Interdisciplinary Reader in 
Philosophy Linguistics and Psychology. 

(~ierzbicka 1969). Anna iVierzbicka. Sncielrania sercmtyczne. 
WrocIaw-W~rs zawa-IQakocat: O~aolinewn, 

(Wilks 1974). Yorj.ak ilk Tatural Language lr~ddrrtandi~~e :::a- 
terns within the AI Paradigm. St n::f ord ;--.tif icial Inte11i.i;~::~: 
Laboratory ..,,?ma :lIZy:-2'7 3, 

(Yiinograd 1 972) . Terry :;'inogred. Undcre tanding Xs taral Iocr~sge, 
Edinburgh: Univeroity Press. 

('Jinogred 1074). Terry Yinograd. Five Lectures sn itificial 
Intelligence Laboratory Xemo XI?: 80.246. Stanford 'Jnl~e~sity. 

(Ziff 1972). Paul Zifr". '?/+"at is said. In (Davidsoz, Iiarnan 
1972) pp 709-723. 
