PRESENT AND FUTURE PARADIGMS 
IN THE AUTOMATIZED TRANSLATION 
"OF NATURAL LANGUAGES. 
Ch. BOITET, P. CHATELIN, P. DAUN FRAGA 
GETA, F-38041 GRENOBLE CEDEX 53X, PRANCE. 
Abstract 
Useful automatized translation must be 
considered in a problem-solving setting, 
composed of a linguistic environment and a 
computer environment. We examine the facets of 
the problem which we believe to be essential, 
and try to give some paradigms along each of 
them. Those facets are the linguistic strategy, 
the programming tools, the treatment of seman- 
tics, the computer environment and the types of 
implementation. 
Introduction 
Machine Translation has been a recurring 
theme ~n applied linguistics and computer 
science since the early fifties. Having not yet 
attained the enviable status of a science, it 
is best considered as an art in the same way as 
Knuth considers computer programming. Failure to 
recognize that MT must be treated in a problem- 
solving setting, that is, as a class of problems 
to be solved in various environments and accor- 
ding to various quality and cost criteria, has 
led and still leads to impassionate, antiscien- 
tific attitudes, ranging polemically between 
dreamy optimism and somber pessimism. Using the 
fairly large body of experience gained since the 
beginning of MT research, we try in this paper 
to extract the most essential facets of the 
problem and to propose some paradigms, alongside 
each of those facets, for usable computer 
systems which should appear in the near - or 
middle - term future. 
As a matter of fact, the phrase 'Machine 
Translation" is nowadays misleading and inade- 
quate. We shall replace it by the more appro- 
priate term "Automatized Translation" (of 
natural languages) and abbreviate it to AT. 
Part I tries to outline the problem 
situations in which AT can be considered. The 
following parts examine the different facets in 
turn. Part II is concerned with the linguistic 
strategy, Part III with the programming tools, 
Part IV with semantics, Part V with the computer 
environment and Part VI with possible types of 
implementation. 
I - Applicability, quality and cost : a problem 
situation. 
1. The past 
Automatized translation systems were first 
envisaged and developed for information 
gathering purposes. 
The output was used by specialists to scan 
through a mass of documents, and, as RADC user 
report shows \[49\], the users were quite satisfied. 
This is no more the case with the growing need 
for the diffusion of information. Here, the 
final output must be a good translation. Second 
generation systems were designed with this goal 
in mind, and with the assumption that good 
enough translations cannot nOW be obtained auto- 
matically on a large scale, but for very res- 
tricted domains (see METEO). Hence, a realistic 
strategy is to try to automate as much as 
possible of the translation proc~s. This is the 
approach taken by GETA, TAUM, LOGOS, PROVOH and 
many others. Here, the problem is to answer 
existing needs by letting man and machine work 
together. 
Another approach comes from AI and is 
best exemplified in \[9\]. Here, the goal is more 
theoretical : how to simulate a human producing 
competent translations ? We will argue that the 
methods developed in this framework are not yet 
candidates for immediate applicability. 
PARADIGM 1 : Future MT systems will be 
AT (automated) systems rather than completeley 
automatic systems. 
2. Applicability 
Automated translation is clearly needed 
by large organizations as the EC or big indus- 
tries having to translate tens or hundreds of 
millions of pages per year. Translations are 
very often urgent, and there is a human impossi- 
bility, as translations needs increase much 
faster than the number of available translators. 
Translators are specialized to certain 
kinds of texts. In the same way, AT systems, 
which are costly to develop and maintain, should 
be tailored to some kind of texts : AT is appli- 
cable when there is a constant flow of very 
homogeneous and repetitive texts, hopefully 
already in machine-readable form. AT should 
allow to integrate automatic rough translation 
and human on- or off- line revision. 
3. Quality 
This is a crucial point, when it comes 
to acceptability by revisors and/or end-users. 
The quality of translation is a very subjective 
notion, relative to the need and knowledge of 
the reader. Traditional counts of grammatical 
errors, false senses,nansenses give only 
indications. 
- 430- 
We believe that quality should be esti- 
mated by the amount of revision work needed, to 
compare it with human (rough) translation, which 
is also very often of poor quality. As errors of 
AT systems are certain to be different from tho~ 
of humans, revisors must have a certain training 
before such a comparison can be made. 
Another measure could be to compare final 
translations, with the same amount of revision. 
We believe this not to be realistic, as cost 
must also be taken into account : translators 
will turn to revision, which is much faster than 
translation, so that they will gain time even if 
they work for revision. 
4. Cost 
The cost of AT should be divided into the 
costs of development, maintenance and use. It is 
of course related to the linguistic and computer 
environments. First, the graph of language-pairs 
should be considered, as development costs for 
an analyzer, say, may be charged to different 
pairs with the same source, of course if a~a- 
lys~ and synthesis are strictly monolingual. 
Easy maintenance calls for sophisticated 
computer systems having an interactive data-b~e 
aspect and concise metalanguages with good, 
incremental compilers. 
Machine time tends to be a minor component 
of the cost of use. Important savings come from 
the integration of the human revision in the AT 
system (see TAUM, LOGOS, GETA), as no further 
typing is required. 
5. Text typology 
AT systems developed for simple texts will 
certainly be less expensive (and probably better) 
than those for complex texts. Let use give a 
tentative hierarchy. The easiest texts are per- 
haps already preedited abstract~ regularly 
entered into data bases. Then come abs~y~a0~, 
which may present surprising difficulties mainly 
due to the tendency to write everything in one 
long and badly constructed sentence. 
Technical documentation, maintenance 
manuals, etc. are more and more written in a 
systematic way, which permits to tailor an AT 
system to their form and their content. See 
however TAUM-AVIATION reports for a sobering 
view on their apparent facility ! Minutes of 
meetings and working document~ may be still 
harder. 
Newspaper articles, even on scientific 
subject matters, tend to accumulate difficult 
or incorrect constructions, and also to jump far 
away from the current subject matter. 
Until AI methods ("third" or "fourth") 
generation are practicable with really large 
data, we don't believe AT systems should even 
try to handle literary, normative or diplomatic 
texts. Revision would just be a new translation. 
PARADIGM 2 : AT systems are now applicable 
only in restricted environments and must be 
tailored to particular 'kinds of texts. 
II - Linguistic strategy 
I. Multilingual or pair-oriented systems ? 
Almost all AT systems divide the process 
of translation in three main logical steps, ana- 
lysis, transfer and synthesis. At one extreme, 
some systems (like METEO) are strongly oriented 
towards a particular pair of languages. This 
means that analysis of the source language is 
performed with the knowledge of the target lan- 
guage. Target lexical units may appear during 
analysis, and syntactic or semantic ambiguities 
in the source are handled contrastively. 
The other extreme is the complete indepen- 
dence of analysis and synthesis. This is the 
approach taken in multilingually oriented sys- 
tems (like ARIANE-78 \[7, 36, 50\], SUSY \[51\], 
SALAT \[18, 20\], TAUM/AVIATION \[33\]). This inde- 
pendence enhances modularity and economically 
justified, as analysis or synthesis are written 
once for each language. Analysis usually repre- 
sents at least 2/3 of the programming effort and 
computing time. 
PARADIGM 3 : We advocate for multilingually 
oriented systems, where the basic software 
itself guarantees independence of analysis and 
synthesis. 
2. What kind of analysis ? 
Should the analysis deliver a structural 
descriptor of the unit of translation, or a 
representation of its meaning, static or dyna- 
mic ? With the first approach, the transfer step 
includes necessarily a lexical transfer and a 
structural transfer. With the second one, the 
result of the analysis is a language-independent 
representation of the unit of translation (sen- 
tence, paragraph(s)). When the lexical units 
themselves are language-free, as in SAM \[9\], we 
call it "pure pivot" approach. When only the 
relations between units are language-free, we 
call it "hybrid pivot" approach (as in the first 
CETA \[34, 35\] system). In the first case, there 
is no transfer, in the second, transfer is 
purely lexical. 
The pivot approach is theoretically very 
elegant. However, past experience with it (on a 
corpus of more than a million words, see 
Vauquois (1975)) shows that it is quite inade- 
quate in real situations, where, very often, 
this representation cannot be obtained, or not 
for all parts of the translation unit. Also, 
human professional translators seem very often 
to produce quite acceptable results without 
actually abstracting the deep meaning and re- 
phrasing it, but rather by using standard syn- 
tactic transformations (like active-passive, 
reordering of nominal groups, passive-impersonal, 
splitting up sentences, etc.) and ... multiple 
431-- 
choice bilingual dictionaries. If deep compre- 
hension fails, it is hence necessary and possible 
to fall back on lower levels of information. 
PARADIGM 4 : The result of analysis should 
be a structural descriptor of the unit of trans- 
lation, where the lexical units are still source 
lexical units and where the linguistic informa- 
tion is "multi-level" : logical relations, syn- 
tactic functions, ~syntactic classes, semantic 
features (all universal for large families of 
languages), and trace information (proper to the 
source language). 
As we argue in Part IV, we don't think the 
result of analysis should include a dynamic 
comprehension of "what is described to happen", 
at least in AT systems for the near future. Let 
us quote Carbonell & al (1978) : "What kind of 
knowledge is needed for the translation of text? 
Consider the task of translating the following 
story about eating in a restaurant...". Unfor- 
tunately, the texts to be translated, as we said 
in Part I, are not stories, but rather abstracts, 
manuals, working documents ... of a very diffe- 
rent nature. 
3. Strategical aspects 
There are some problems the analysis writer 
can not escape. Should problems such as ambigui- 
ties be solved as soon as they appear, or not be 
solved altogether, or is it better to devise 
strategies to decide as late as possible, or 
more complex heuristics ? 
PARADIGM 5 : AT systems to be developed in 
the near future should allow complex linguistic 
heuristics. That is, we feel that preferences 
computed by the use of weights derived from some 
frequency counts are not enough, and that lin- 
guists should program what they see as being 
essentially heuristic in the linguistic pro- 
cesses. Hence further requirements on the pro- 
gramming tools, which should at least include 
such control structures as controlled non- 
determinism. 
III- Programming tools : algorithmic models and 
metalanguages 
I. History 
The first MT researchers programmed direc- 
tly in machine language. Until now, SYSTRAN 
systems are essentially made of programs and 
tables written in IBM 370 macroassembler. Then 
came systems based on a simple formal model, 
like context-free grammars and Q-systems. These 
systems rely on a general algorithm over which 
the rules have no control. Systems allowing such 
controls (PROLOG \[52\], ATEF \[14, 15\], ROBRA \[50\], 
ATNs \[47\] and derived models like REZO \[32\], 
PLATO, DEDUKT \[18, 20\]) were created in the 
seventies. 
Now, the programming languages used to 
write the linguistic part of AT systems include 
usual programming languages such as macro- 
assembler, FORTRAN, ALGOL, PL/I, LISP, as well 
as specialized languages (see above). 
2. The need for powerful data and control 
structures 
In our view, usual programming languages 
are inadequate as metalanguages to be used for 
writing the linguistic data and procedures in an 
AT system. 
PARADIGM 6 : Adequate metalanguages should 
include built-in complex data-types such as 
decorated trees and graphs as well as control 
structures for non-deterministic, parallel and 
heuristi c programming. 
Note that parallelism may be of two diffe- 
rent kinds : processors working independently on 
independent data structures and processors wor- 
king on a common data structure (e.g. a normal 
context-sensitive grammar is not equivalent to 
the same grammar used in parallel, see 
S ~ AB, A ~ a/-B, B ÷ b/A-). Many recent specia- 
lized programming languages include a form of 
non-determinism, but very few have parallelism 
(ROBRA) or control functions for heuristics 
(PROLOG, ATEF, REZO). 
Of course, these metalanguages should 
include more classical control structures such 
as iteration, recursion or selection. Note that 
dictionaries are simply big "select" constructs, 
possibly non-deterministic (one-many). 
3. Complexity, decidability, adequacy 
If one takes all necessary data-types with 
all possible operators and all control struc- 
tures, the model obtained is very likely to have 
the (maximal) computing power of a Turing machi- 
ne. Hence, no general bound or estimate for the 
dynamic complexity of programs written in that 
formalism may be given. On the other hand, as 
all we want to program in AT systems is cer- 
tainly sdbrecursive, another approach is to 
define several subrecursive algorithmic models 
with associated known (or studyable) complexity 
classes. This was the original approach at GETA, 
with the ATEF, ROBRA, TRANSF and SYGMOR algo- 
rithmic models, designed to be decidable and of 
linear complexity. 
As a matter of fact, decidability is a 
very practical requirement. However, general 
constraints imposed to guarantee decidability 
may make certain things unnecessarily clumsy to 
write. Perhaps a better idea (implemented in 
ATNs, ATEF and ROBRA) is to build algorithmic 
models as extensions of decidable models, in 
such a manner that sources of undecidability are 
easy to locate, so that particular decidability 
proofs may be looked for. For example, the fun- 
damental operator of ROBRA is the parallel 
application of the rules of a transformational 
grammar to an object tree. 
432- 
Normal iteration of this operator must termi- 
nate, due to some marking mechanism. However, a 
grammar in "free" iteration mode may never 
terminate. 
Last, but not least, these algorithmic 
models must be adequate, in the sense of ease 
and concision of writing. We sum up with 
PARADIGM 7 : The complex operators asso- 
ciated with the data types should be adequate, 
their complexity (time and space) should be 
reasonably bounded (O(n) to O(n3) ?) and there 
should be decidable underlying algorithmic 
models, so that so'urces of undecidability could 
easily be traced. 
IV - Semantics 
i. Two different notions 
Semantics are understood differently in 
linguistics, logic and computer science. In the 
latter, attention is focused on the ways of 
expressing data and processes. A system is said 
to be "syntactic" if it operates within the 
framework of formal language theory, that is by 
combinatorial processes on classes or "features'~ 
In a "static" semantic system, there is a 
fixed model of some universe, possibly repre- 
sented as a thesaurus, or as a set of formulae 
in some logic, on which a formal language is 
interpreted. 
A system incorporates "dynamic semantics", 
or "pragmatics", if the interpretation of the 
data it processes may alter the model of the 
universe, or create a parallel model of some 
"situation" in this universe. 
2. Classical approaches 
Existing AT systems of reasonable size, 
that is incorporating several thousands of lexi- 
cal units and quite exhaustive grammars, rely 
essentially on semantics by features. They may 
be quite refined and specialized to a domain 
(e.g. METEO), and, in that case, this method 
may give surprisingly good results. 
Although the basic softwares allows to 
relate lexical units by using (monolingual or 
bilingual) dictionaries, this possibility is 
hardly used in the current applications at TAUM 
see TAUM/AVIATION) or at GETA (see \[50\]). For 
instance, classical relations such as antonymy, 
generalization, particularization are not coded 
in the dictionaries. 
3. AI proposals 
AI proposals fall into two classes. The 
first refers essentially to static semantics, 
and may be illustrated by Wilks' "preference 
semantics" \[37-44\] or Simmons "semantic net- 
works" \[30\]. 
As applied to AT, these methods have only been 
incorporated in very small size test programs 
incorporating at most some hundreds lexical 
units. However, we feel that their simplicity 
and relative economy in coding effort make them 
usable in near-term AT systems, under the essen- 
tial condition that, as in Wilks' model, it is 
not necessary to code completely every lexical 
unit, and that the associated computing effort 
is controlled by the linguist and undertaken 
only when necessary, that is when a problem 
(like ambiguity or anaphoric reference) has not 
been solved by s~mpler means. 
The second class of AI proposals relates 
to dynamic semantics and originates in the 
"frames" proposed by Minsky \[12\], and now pro- 
posed by other teams as "scripts", "plans" or 
"goals" \[9, 27-29\]. They are certainly very 
attractive, but have been demonstrated on very 
particular and very small size situations. 
As we said above, texts to be translated 
with AT systems are more likely to be technical 
documents, abstracts, instructions for use, main- 
tenance manuals, etc., than stories about res- 
taurants or earthquakes. Each text doesn't rely 
on one clear-cut "script", or "type of situa- 
tion", known a priori. Rather, such texts very 
often don't describe situations (see a computer 
manual), or, at the other extreme, their content 
might be understood as ... the description of 
hundreds of scripts (see aviation maintenance 
manuals). 
Hence, our objection to the use of such 
methods is twofold. First, the coding effort, 
in principle, would be enormous. Charniak's 
frame for painting \[;2\], although admittedly 
incomplete, is 19 pages of listing long (in a 
high-level language !), and we suppose he spent 
rather a long time on it. Just think of what it 
would cost to code 5000 basic frames, which we 
believe would be reasonable for, say, the 
domain of computers. Second, if the texts des- 
cribe types of situations, then it is necessary 
to understand these texts in order to code the 
necessary scripts, which will be used ... to 
understand the text again ! 
This circularity has two consequences. 
First, only very general scripts might be hu- 
manly coded by using general previous knowledge 
about the domain. Second, if we want to use such 
methods extensively and at levels of detail at 
which they begin to give better results than 
simple approaches, then AI researchers should 
provide methods for the automatic extraction of 
scripts or frames from large bodies of texts, in 
an efficient (and perhaps interactive) way. That 
is, the use of such methods on wide domains and 
large amounts of texts entails automatic lear- 
ning. Another problem is to automatically find 
which script is relevant to the current portion 
or text. 
--433-- 
4. Concluding remarks 
As in other problem-solving situations, 
simple methods should continue to be used in 
conjunction with more sophisticated ones. 
Unfortunately, proponents of "very high" seman- 
tics seem too often to concentrate on interes- 
ting high level phenomenaeas anaphoric reference, 
discourse structure, causality and reasoning and 
to forget at the same time persisting and very 
frequent lower-level difficulties such as ambi- 
guity of prepositional group dependency. Take 
Riesbeck's \[25 p. ll\] example ,'John hurt Mary 
because Mary informed Bill that John advised 
Rita to prevent Bill from buying the book by 
giving the look to John". It seems obvious to 
the author that "by giving" relates to "prevent". 
However, it could also relate to "hurt", 
"informed" and "advised" ("buying" being exclu- 
ded because of the coincidence of reference due 
to "the"). Take also Vauquois' example "the 
minister spoke of the candidate to the presi- 
dency", and many other occurring with conjunc- 
tion and enumeration. 
PARADIGM 8 : Reasonably large scale AT 
systems can rely only On semantics by features 
and static semantics in the near future. Scrip__t 
like methods must be complete d by automatic 
script generation and retrieval procedures 
before they can be used extensively. Semantic 
methods must complete and not discord previous 
ones. 
V - Computer environment for the users 
I. Essential functions and types of users 
The are different kinds of users of AT 
systems, intervening in different ways for dif- 
ferent purposes, related to the functions of 
creation, maintenance and use. Specializgd 
linguists create the linguistic systems or 
subsystems, lexicographs and terminologists 
create and update the dictionaries, revisers 
and translaters use the system as a tool to 
produce translations, and the end user wants to 
know nothing about it, but is nevertheless the 
final judge. 
PARADIGM 9 : A modern AT system must 
allow large degrees of interactivity at ail 
functional levels, be transparent to the user, 
contain a (possibly specialized) data-base 
management system (for handling grammars, dic- 
tionaries, texts and their different versions 
as well as intermediate results and statistical 
information) and be integrated (from preediting 
and/or checking to revision and (photo) 
composition). 
2. Types of use 
At creation time, interactivity is cer- 
tainly essential, even during parts which will 
be fully automatic in a production environment. 
Perhaps a mode should be provided in which the 
system could ask simple questions (choice of 
equivalents, for instance) to a translator 
sitting at a screen while doing the automatic 
rough translation part. Even that may be too 
costly in a production environment. 
For maintenance and updating of the lin- 
guistic data, we believe it is essential that an 
AT system provides ways of feed-back and commu- 
nication between the different kinds of users, 
and between users and the system. 
3. Human aspects 
The human and social aspects should not be 
neglected. To force a rigid system on revisors 
and translators is a guarantee of failure. It 
must be realized that AT can only be introduced 
step by step into some preexisting organizatio- 
nal structure. The translators and revisors of 
the EC did not only reject Systran because of 
its poor quality but also because they felt 
themselves becoming "slaves of the machine", and 
condemned to a repetitive and frustrating kind 
of work. 
PARADIGM I0 : AT systems must be such that 
the users keep control over them, and not vice 
versa. 
VI - Types of implementation 
This section is voluntarily short, as this 
question is not particular to AT systems. Hard- 
ware requirements for large MT systems are 
already met by minicomputers like IBM's 4300 
series. Software requirements such as time- 
sharing and virtual memory are also available. 
Typically, GETA's current prototype Russian- 
French translation system executes with a cen- 
tral memory less than 1,5 Mbyte without any 
disk-access during the translation process, and 
uses 12 Mbytes on disk for linguistic files and 
the text data-bases. If the dictionaries would 
increase in size to, say, 40000 lexical units 
(more general than words, or roots), than 
3 Mbyt~of virtual memory and 20 to 25 Mbytes 
on disk would be needed. Even microcomputers 
might support such systems in the (longer term) 
future. 
For the time being, such systems may be 
centralized, and operate on big computers, or 
be distributed on minicomputers, possibly lin- 
ked through a network. The machine may be dedi- 
cated or not, and so forth. In addition to the 
hardware side, the software side is also impor- 
tant. Portability and efficiency are often con- 
flicting goals. The implementation language(s) 
(~0£ the metalanguages) may be low-level 
(assembler, LP language) or high-level (FORTRAN, 
PASCAL, PL/I, ALGOL68, LISP, ADA,...). Another 
possibility is to devise a special abstract 
machine for the metalanguage. 
434 
We believe that only the last two solutions 
should be considered, with re~£ portability and 
efficiency as the main criterion for choosing a 
high-level language. 
As a likely development, we foresee the use 
of AT systems first on big computers in large 
organizations, with or without teleprocessing, 
and then, in bureautics, on local minicomputers. 
However, some recent experience indicates that 
local development of user-tailored applications 
may well be done before bureaucratic inertia in 
large organizations allows decisions to be taken. 
Acknowledgments 
We would llke to thank Pr. Vauquois as a 
principal inspirator of this work, although the 
errors are of course ours, and although his ideas 
on some points are certainly more advanced than 
those exposed here in the framework of possible 
near-immediate large scale applications. 
Note : EC 
AI 
European Community 
Artificial Intelligence 
--436- 

BIBLIOGRAPHY 

\[I\] Y. Bar-Hillel (1960), "The present status of 
Automatic Translation of languages" in : 
Advances in Computers, VI, 91-163. 

\[2\] Y. Bar-Hillel (1967), "Die Zukunft der 
maschinellen Uebersetzung, oder : warum 
Maschinen das Uebersetzen nicht erlernen", 
Sprache im technischen Zeitalter, 23. 

\[3\] Y. Bar-Hillel (1971), "Some reflections on 
the present outlook for High Quality Machine 
Translation", in : Feasibility study on fully 
automatic high quality translation, (RADC) 
Univ. of Texas. 

\[4\] Ch. Boitet (1976), "M6thodes s~mantiques en 
TA", TA-informations n ° l, 3-42. 

\[5\] Ch. Boitet (1976), "Un essai de r~ponse 
quelques questions th~oriques et pratiques 
li6es ~ la traduction automatique. D6fini- 
tion d'un syst~me prototype". Th~se d'Etat, 
Univ. de Grenoble. 

\[6\] Ch. Boitet (1976), "Probl~mes actuels en 
traduction automatique. Un essai de r~ponse", 
COLING-76, Ottawa, Preprint n ° 33. 

\[7\] Ch. Boitet (1977), red., "O~ enest le GETA 
d6but 1977 ?", Comm. groupie, 3~me congrgs 
europ~en "Franchir la barri~re linguistique", 
Luxembourg, p. 599-636. Also available in 
English and German. 

Ch. Boitet (1977), "MT and the problem of 
understanding natural language", colloque 
franco-sovi~tique sur la TA, Moscou, 
dgc. 1977. 

J. Carbonnel, R.E. Cullingford, 
A.V. Gershman (1978), "Knowledge based 
Machine Translation, RR ~146, AI Project, 
Yale Univ., and Proc. COLING-78, Bergen. 

E. Charniak (1975), "A partial taxonomy of 
knowledge about actions", ISSCO, WP ~13, 
Castagnola. 

E. Charniak (1975), "Organization and infe- 
rence in a frame-like system of common 
sense knowledge", ISSCO, WP~I4, Castagnola. 

E. Charniak (1978), "A framed painting- 
representation of a common-sense knowledge 
fragment", Cognitive Science, V2, Nl. 

J. Chauch~ (1974), "Transducteurs et arbo- 
rescences. Etude et rgalisation de syst~mes 
appliqugs aux grammaires transformation- 
nelles". Th~se d'Etat, Univ. de Grenoble. 

J. Chauch~ (1975), "The ATEF and CETA 
systems", TA-informations n ° 2 & AJCL, 
Microfiche 17, 21-39. 

J. Chauch~, P. Guillaume, M. Qu~zel- 
Ambrunaz (1973), "Le syst~me ATEF", 
Doc. GETA G-2600-A, Univ. de Grenoble. 

A. Colmerauer (1971), "Les syst~mes-Q, ou 
un formalisme pour analyser et synth~tiser 
des phrases sur ordinateur", in : TAUM-71, 
Univ. de Montreal. 

R.W. Floyd (1967), "Non-deterministic 
algorithms", JACM, VI4, N4, 636-644. 

C. Hauenschild, E. Huekert, R. Maier (1978), 
"SALAT : Machine Translation via semantic 
representation", in : B~uerle & al. 

Th. Hoffman (1978), "Semantics in aid of 
automatic translation", COLING-78, Bergen. 

E. Huckert (1979), "Automatische Synthese 
des Franz~sischen aus einer logischen 
Basis", AQ-Verlag, Dudweiler, Saarland. 

R. Kittredge, L. Bourbeau, P. Isabelle 
(1978), "Design and implementation of an 
English-French Transfer Grammar", COLING-~, 
Bergen. 

N. Nedobejkine (1976), "Niveaux d'interpr~- 
tation dans une traduction multilingue : 
application g l'analyse du russe", 
COLING-76, Ottawa. 

R. Quillian (1968), "Semantic memory", in : 
Semantic information processing, MIT Press, 
216-270. 

R. Quillian (1969), "The teachable lan- 
guage comprehender : a simulation program 
and a theory of language", CACM, V12, N8, 
459-476. 

C.K. Riesbeck (1974), "Computational 
understanding : analysis of sentences and 
context", ISSCO, WP~4, Castagnola, 241 p. 

E. Sandewall (1971), "Representing natural 
language information in predicate calculus", 
in : Machine Intelligence 6, Meltzer & 
Mitchie, ed., American Elsevier. 

R.C. Shank (\]973), "Identification of 
conceptualizations underlying natural lan- 
guage", in \[48\], 187-247. 

R.C. Shank (1974), "Understanding para- 
graphs", ISSCO, WP~6, Castagnola. 

R.C. Shank, C.J. Rieger III (1974), 
"Inference and the computer understanding 
of natural languages", Artificial intel- 
ligence 5, 373-412. 

R.F. Simmons (1973), "Semantic networks : 
their computation and use for understanding 
english sentences", in \[48\], 63-113. 

R.F. Simmons, J. Slocum (1972), "Generation 
of english discourse from semantic net- 
works", CACM, VI5, NIO, 89\]-905. 

G. Stewart (1975), "Le langage de program- 
mation REZO", TAUM, Univ. de Montreal. 

TAUM (1979), "Presentation de la cha~ne de 
traduction informatis~e TAUM/AVIATION", 
Univ. de Montreal, 27 mars 1979. 

B. Vauquois (1968), "Structures profondes 
et traduction automatique. Le syst~me du 
CETA", Revue Roumaine de linguistique, 
13, 1057130. 

B. Vauquois (1975), "La traduction 
automatique ~ Grenoble", Doc. de Linguis- 
tique Quantitative n ° 24, Dunod, 184 p. 

B. Vauquois (1976), "Automatic translation- 
a survey of different approaches", 
COLING-76, Ottawa, and SMIL I, 127-135. 

Y. Wilks (1968), "On-line semantic ana- 
lysis of english texts", Mechanical 
Translation, V|;, N3 & 4, 59-72. 

Y. Wilks (1972), "Grammar, meaning and the 
machine analysis of language", Routledge 
London, 198 p. 

Y. Wilks (1973), "An artificial intelli- 
gence approach to machine translation", 
in \[48\], 114-151. 

Y. Wilks (1973), "Preference semantics", 
Stanford AI lab., AIM-206, CS-73-377. 

Y. Wilks (1975), "An intelligent analyzer 
and understander of english", CACM, V\]8, 
N5, 264-274. 

Y. Wilks (1975), "Seven theses on artifi- 
cial intelligence and natural language", 
ISSCO, WP~17, Castagnola. 

Y. Wilks (1976), "A preferential, pattern- 
seeking, semantics for natural language 
inference", Artificial Intelligence 6. 

Y. Wilks & M. King (1976), "Semantics, 
Preference and Inference. A full descrip- 
tion of a system and a program", ISSCO, 
WP~18, Geneva. 

T. Winograd (1971), "Procedures as a repre- 
sentation for data in a computer program 
for understanding natural language", 
AI-TR-17, MIT. 

T. Winograd (1973), "Procedural model of 
language understanding", in \[48\],152-186. 

W.A. Woods (1975), "Syntax, semantics and 
speech", BBN report n o 3067, AI report n°27. 

Shank & Colby, ed. (1973), "Computer models 
of thought and language", Freeman & Co, 
San Francisco. 

Z.L. Pankowicz, technical evaluator (1973), 
"User's evaluation of machine translation 
Georgetown MT system (1963-1973)", 

RADC-TR-73-239, Univ. of Texas at Austin. 
C. Boitet, N. N~dobejkine (1980), "Russian 
French at GETA : outline of the method and 
detailed example", to appear in the Proc. 
of COLING-80, Tokyo. 

H.D. Maas (1978), "Das Saarbr~eker Ueber- 
setzungssystem SUSY", Sprache und Datenve- 
rarbeitung, 43-62. 

A. Colmerauer, H. Kanoui, M. Van Caneghem 
(1979), "Etude et r~alisation d'un syst~me 
PROLOG", Groupe d'Intelligence Artificielle, 
Univ. d'Aix-Marseille II. 
