Англомовні навчальні матеріали для набору та редагування

1.2. Ambiguity Problem in Translation.

I hope I managed to persuade you that the formalization of translation is indeed a difficult task. Now we shall discuss why this problem is so hard to resolve.

The answer is contained in one word: "ambiguity".

Ambiguity is the property of language units to bear several dif­ferent meanings.

It is a fact widely known and you, of course, met it more than once. Some languages, like English or Chinese are more ambigu­ous, others like Ukrainian or German are less so.

It depends on how a meaning is expressed in one or another lan­guage.

Let us look at some examples:

• "Ви бачите голову?" In this Ukrainian sentence the ambigu­ous word is "голова "(a chairperson and a head).

« Шел дождь и два студента, один в пальто, другой - в институт". In this Russian example we observe two cases of ambi­guity - that of "шел" (It was raining vs. Students were walking) and that of the preposition "в"(In vs. To).

• In another Russian example, Boulat Okoudzhava's line "Какие женщиньї на нас бросают взорьі... "the ambiguity is ex­pressed in the relation of the attribute "какие" to the nouns "женщиньї" and "взорьі" (What women cast looks or What looks do the women cast?).

• The English sentence "A bare conductor ran in the car" (В ва­гоне проложен неизолированньїй провод) is so ambiguous that it has become a translators' joke (Гольїй кондуктор бегал по вагону

In the above examples you should note that not only is the meaning of the words ambiguous but also the relations between them.

We shall come to it again later. Just remember now that the units of all languages possess some degree of ambiguity. And as it has already become common with us let us ask ourselves: "Why?" Why are language units ambiguous?In order to answer this question let me first tell you that any language unit consists of these three indivisible components:

1. a sign (form, expression) of a language unit

(e. g. book, it,- tion, a, etc.);

2. a concept (meaning) of a language unit (the mental con­tent of the language unit conventionally related to the sign in the minds of language speakers);

3. a denotatum (a fragment of the real world, including the inner world of human beings, that corresponds to a given con­cept)

According to common scientific opinion there may never be a di­rect link between the sign and the denotatum. Never ever. Which is shown by a dashed line in the famous triangle of Ogden and Rich­ards (Fig 1).

CONCEPT

SIGN DENOTATUM

FIG 1

Why does this situation make the words and the relations be­tween them so ambiguous? The answer is straight-forward enough: because there is no direct link between an image of the real world and the form of a word, that is, others simply do not see what you mean.

Now, let us ask ourselves if the meaning of our speech is really that unclear. You would, perhaps, say: "No.

You understand me; I understand you; we understand others if we all speak one and the same language."

So, on the one hand, words are ambiguous and there are lots of examples (those recently discussed are convincing enough), but on the other, when speaking a language, you make others understand you and you understand others who speak this language. It looks like a contradiction, a paradox, but it is not, and to explain it let us look at some history. Among the hieroglyphs one can find, for ex­ample, an image of a an eye (Fig.2).

This eye image may be primitive but it is, indeed, an eye mean­ing "to see". In this and some other hieroglyphs we do observe a similarity between the sign (hieroglyph) and a fragment of the real word (an eye).

So, at the early stages of language development there was a direct link between the language signs and the real world. This direct con­nection was observed also in spoken, rather than only in written language expressions. It may be found even in modern languages, though rather seldom. For example the name of the cuckoo bird is derived from the "cuckoo" sound typical for this bird.

Then in the course of language evolution similarity was replaced by convention.

Instead of designing or finding word images similar to those in the real world, people simply agreed that such and such expression will have such and such a meaning.

It is similar to mathematics, where if you have, say, symbol "x" in an equation, you must specify that "x", for instance, is some in­tegral number.

When you learn a language, you join a convention. And when you speak you are understood only by the members of this conven­tion, that is by the speakers of this language.

For example, if you understand the word "marzyciel'you are a member of the Polish language convention, if not, then you are not.

Then, if the meanings of the language units have been agreed upon, why do we still have ambiguous meanings? The answer to this question is not that simple and in a way it is beyond our discipline, and more in the realm of historical linguistics to ex­plain. We may only guess that the interpretations of the agreed meanings change with time (our ancestors understood some words differently), that individual interpretations and those of certain groups of speakers are different, and one and the same word acquires new meanings under the impact of these and other factors.

In short, languages are developing and in this process words develop new meanings.

As a result we have what we have ambiguity. And our task is to get rid of the ambiguity in the process of language formalization. This task is called disambiguation.

In natural language processing disambiguation is a process of meaning clarification. During disambiguation the meaning is clasified using formal methods.

When we use a language (i.e. speak, listen, read, or write) the disambiguation is performed continuously at the intuitive level. We may say that:

We understand other people speaking the same language as well as they understand us because every time we speak or listen to others speaking, write or read we clarify the meaning of the words using the following three powerful disambiguation tools:

• context environment,

• situation and

• background information.

For instance, if we take the ambiguous examples we discussed above in a broader context they will become quite clear:

1. "На зборах мене спитали: Ви бачите голову?".

2. "Какие женщиньї на нас бросают одобрительние взорьі

3. "A bare copper conductor ran in the car"

In all of these examples the context elements shown in bold type clarify the meaning of ambiguous words.

Situation is another factor that makes the meaning of the am­biguous words clear.

For example, if somebody asks you at a meeting: "Ви бачите голову?" you will undoubtedly understand that what was meant had been a chairperson rather than a head - the situation of the meeting prompts you for the correct decision.

If you are discussing, say, an electrical installation, you will, perhaps, have no doubt that a wire rather than a person is under­stood by the word "conductor".

However, neither the context nor the situation alone will lead you to the correct answer, unless you already have the relevant background information.

The background information is your common sense, your knowl­edge of the way the things are in life.

For example, you will understand the sentence "The cranes are flying" as "Летят журавли" because from your life experience or from books you know that the lifting machines (cranes) can not fly, or you will understand the phrase "Heavy powerful cranes" as "Тяжельїе мощньїе краньї", because your experience suggests to you that the birds are less likely to be described as "heavy" or "powerful".

And it is more likely that you will understand "There are cranes in the port" as "Краньї в порту", though theoretically there may be birds (cranes) as well.

Now, coming back to the question, why we understand each other in spite of all the ambiguity of the language, we may answer it by saying: "We understand people speaking the same language, we can read, write, and translate due to continuous analysis of the con­text and situation and the appropriate use of the relevant background information."

So, it's as simple as that. And it is rather simple for us because we possess a powerful and quick-acting system of information ana­lysis which has no known automaton.

Partly this system is based on logic but the larger part of it is what we call intuition.

I dare say that intuition is subconscious logic. That is we sub­consciously perform very quick logical operations, not even know­ing about this ourselves.

Let us take an example.

There is a fragment of a poster with three letters "ИДА" on the wall of an opera-house. I think you will agree with me, if I say that any educated person would understand it as a fragment of a poster of Verdi's "Aida" rather than the woman's name "Ida" or "Lida". Why? Because of the following main factors:

1. You know about the existence of an opera called "Aida" (you may have watched it or read or heard about it).

2. You know that the building you see is an opera house (you may have already seen it or you may recognize the pe­culiar architecture of an opera house).

3. You know that the paper you see is a fragment of a poster, rather than just a piece of writing paper (you may judge by the texture of this paper or on the type of the print, etc.).

In other words, you make a very fast factor analysis being to­tally unaware of the process yourself.

However, in order to simulate this analysis and make a com­puter or any other robot "understand" it you have to have:

• a list of all factors involved;

• an interrelated structure of those factors and

• weights (ratings) for each factor.

Let's look at a simple example of factor analysis for a hypotheti­cal translation system. The task is to find factors that would allow this system to choose one of the three translation alternatives for the sentence:

"The walls of the vessel appeared too thin to withstand such pressure"

1. Стенки сосуда оказались слишком тонкими для того, чтобьівьідержать такоедавление.

2. Стенки емкости оказались слишком тонкими для того, чтобьі вьщержать такое давление.

3. Стенки корабля оказались слишком тонкими для того, чтобьі вьщержать такое давление.

Our assumption will be that semantically related words in the context environment are the factors which determine the selection of a translation alternative, namely:

If in the context environment we find words "blood", "coro­nary", "patient", "medical" and the like then the translation of "vessel" is "cocyd ".

If in the context environment we find words "storage", "proces­sing", "vent", "chemical"and the like then the translation of "vessel" is "емкость".

If in the context environment we find words "sea", "ship", "board" "cargo" and the like then the translation of "vessel" is "корабль ".

This method is called semantic network analysis and it sometimes gives fairly good results.

Thus, it seems that for the time being I have presented to you enough evidence to prove that the problem of formal disambigua­tion is a difficult one indeed.

Let us now discuss briefly the varieties of ambiguity you will face in the process of translation formalization.

The indirect relationship between the word and its denotatum is one of the factors that make a language ambiguous and result in polysemy of language units.

In its broader sense polysemy is the ability of a single word or expression to have different meanings.

It is probably not necessary to explain the notion of polysemy in detail, because it is well known, especially to those who deal with the English language. It is enough to look at the words "get or" board"in a dictionary to prove the polysemantic nature of English and similar analytical languages. Again it is evident that polysemy is an obstacle in the way of translation.

Also at this stage we shall not discuss the phenomenon of homonymy, the cases of which are known too well to constitute a serious translation problem (e. g. "crane" (a bird) and "crane" (a machine) in the example given above).

The concept corresponding to a word or expression is poorly de­fined in relation to its denotatum within a given language and it becomes even fuzzier when two different languages are matched in translation.

The concept of a denotatum is sometimes different in different languages.

Thus, L. Zgusta reports that there are only three colors in some African languages, and four in some languages in Northern Brazil as opposed to the seven basic colors we Europeans are used to.

For the speakers of a language the whole extralinguistic world, either material or not, is organized into concepts and the differ­ences in their conceptual organization are rather frequent for any pair of languages.

These differences are one of the obstacles one has to cope with when learning a foreign language or when making a translation.

To the above difficulty for the translator one should also add the problem related to the perception of the extralinguistic world by various social, professional, and other groups.

For instance, compare the conceptual content of the word "bar" in the layman's mind and in that of a practicing lawyer.

Apart from these discrepancies in perception, translation is also hampered by such an innate feature of a language as connotation.

According to L. Zgusta connotation consists of all the compo­nents of a meaning that add some contrastive value to the basic con­cept of a word.

This value may be positive (e. g. "eagle" - a connotation of free­dom, courage) or negative (e. g. "fox" - cunning, deceptive).

Connotations may be related to the national (or religious) pecu­liarities, for example in Moslem world a dog has clearly negative connotations as opposed to positive ones for the Europeans (a friend of a man, clever, devoted, etc.).

So, polysemy and connotation are features of a language which hamper the solution of the translation problem.

Whereas context environment, situation and background information make the transla­tion possible (at least, to some extent).

In other words, in the translation process we observe continuous interaction of the factors causing ambiguity, on the one hand, and those forming the basis of disambiguation, on the other (Fig. 3).

Factors Ambiguity

individual/ national perception

polysemy

connotation

TRANSLATION PROBLEM

context

situation

background information

Disambiguation Factors

1. 3. Theory, Model, Algorithm - Differences and Common Features

In this book we shall discuss theories, models, and algorithms of translation and it seems natural to begin with the general defini­tions of these terms.

So, what is a theory, to start with?

Theory is the broadest and the most general of the above terms.

Theory is defined as "... a structure in which each step depends on preceding steps...The whole structure rests upon observation and theoretical assumptions"

A theory is used for description, explanation, and classification of observed events or for the prediction of future events. A theory is always a regularity rather than an individual fact. A theory is based on known laws and provides guidance for the discovery or formulation of other (unknown) laws.

This means that for a theory you will need, first of all, the ob­served events or observables (i.e. something that is registered by human perception organs or is deducible from what has been per­ceived) and a description of each observed event.

As a next step one may design a structural classification of the observables showing their interrelation and interdependence, which will then serve as a theory explaining some current relationship be­tween the observables and predicting future behavior of the latter.

• Identify all observed events (both already known and new);

• explain their meaning (formulate the concepts of the events);

• establish interrelation and interdependence of the events;

• predict future events and you will have a theory

In other words, a theory is a system of generalized knowledge in one or another area of human activity.

In a theory one may distinguish between the formal (repre­sentational) and conceptual (interpretational) parts.

In such natural sciences as physics or chemistry the formal part of the theories is represented by formulas and equations and the conceptual part consists of their interpretations.

This distinction becomes more and more often the case for lin­guistics. Although previously in linguistics the formal part was rarely represented by mathematical expressions, nowadays with the development of such disciplines as structural, mathematical, and computational linguistics it is observed rather often.

It is worthwhile to demonstrate the structure of a theory using an example from natural sciences:

The following steps were necessary to build the Kinetic Theory of Gases:

• Empirical foundation. Observation of pressure, temperature, and mass of different gas samples

(Identification of the observables).

• Formulation. Formulation of the concepts of pressure, tem­perature, and mass.

• Introduction of the known concepts. Introduction of the con­cepts of a molecule, momentum, force, and energy.

• Final statement. Formulation of the relationship between the observed and the known concepts to explain the behavior of gases under different pressures and temperatures.

Any theory usually begins with the formulation of a hypothesis. A hypothesis may be considered as an unconfirmed predecessor of a theory.

A hypothesis or an interrelated set of hypotheses become a the­ory when confirmed empirically.

Hypothesis is a scientifically proved assumption either about an event which cannot be directly observed or about a regularity ex­plaining the behavior of a known set of events.

Generally, a theoretical study involves such a number of hy­potheses that only a small part of them later becomes true scientific knowledge, i.e. a theory.

Let us discuss the hypothesis and the theory using an example of the linguistic study, involving both formal and conceptual parts.

It is universally known that statistical methods are the most widely used ones to check the empirical correctness of a hypothesis.

Thus, in the early days of statistical linguistics, a binomial or Poisson's distribution was used to study the behavior of discrete linguistic units (syllables, affixes, words, etc.) in speech.

The basic assumption (hypothesis) in this case was that speech had been a simple flow of independent (random) linguistic events.

However, this approach, being a very rough and oversimplified approximation, appeared to be of limited use in linguistic study.

Then it was correctly presumed that the formation of words, word combinations, sentences, and utterances had been guided by the interaction of both random and predetermined processes - that is from a linguistic viewpoint there are two types of linguistic units in the flow of speech:

• those involved in the description of situations objectively ex­isting independently from a language (random events) and

• those units which represent the rules of the system and the norm of a language (predetermined events).

For example, in the sentence "Я буду ждать" the word "я" is random, whereas the forms of the words "буду" and "ждать" are predetermined by the relevant grammar rules (i. e. you cannon use the forms "бьіть" or "ждет" instead of "буду" and "ждать" because of the agreement between Personal Pronoun as Subject and Verbal Predicate and Future Tense formation rules).

This new, more accurate approach to the interpretation of speech generation (new hypothesis) replaced previously existing ones and the Chebanov-Fuchs distribution representing the new interpretation was substituted for the binomial or Poisson's one.

It is not worth showing here the curves or analytical formulas of both distributions involved in the case. This would need a detailed explanation not related directly to the subject at hand.

It is more important to stress that this example provides a clear distinction between a theory and a model - linguistic explanations of language unit co-occurrence are theories, whereas the Poisson's and Chebanov-Fuchs' distributions are statistical models.

This brings us to the definition of a model.

A model is an abstract logical structure which describes an ob­ject or a system by analogy and is intended for the explanation of their behavior. The model is a structure outside the object or system being modeled

In the most general sense a model is a physical device which represents an object. An analogue model represents an analogy with the object, for example, diffusion is often described using a swarm of bees as an analogy.

A theoretical model describes an object or a system by using ob­servables, the behavior of which explains various properties shown by the object or system.

Usually a theoretical model uses an analogy of an object or a system. As it may be seen from the comparison of the two, a model is a more exact and specific description of an object or a system than a theory.

Moreover, a theory usually establishes a system within an object or phenomenon, whereas a model just describes this system. Lan­guage models form a subcategory of theoretical models.

Language modeling is a method of formulation of certain gen­eral hypotheses about the structure of a language as an abstract semiotic system as well as a method of experimental evaluation of the correctness of these hypotheses.

Due to the discrete nature of language units a language model is an orderly sequence of discrete elements. The number of elements is usually finite.

There are two approaches to any modeling (including language and translation modeling) - modeling by deduction and by induc­tion.

Deduction is formulation of certain sequences of abstract sche­mes being more or less close approximations of real facts and phe­nomena of a language and evaluation of the correctness of these schemes by application to real language structures. Models pro­duced by deduction are called synthetic or generative.

For example, let us make an assumption that the sequence of syntactic classes consisting of an Attribute (A), Noun (N), Verb (V) and Verb Modifier (M) reflects the syntactic structure of a certain number of English sentences

In this case

S=A, N, V, M

will be a syntactic model of an English sentence obtained by de­duction.

Then let us check the correctness of this model substituting the elements of the model for the real English words of the corre­sponding classes. Such examples as "Small children cry often", "New cars run faster", "Healthy people sleep well", and the like prove that our original assumption was correct andS=A, N, V, M is, indeed, a model of the syntactic structure characteristic of a cer­tain set of English sentences.

The method of model substitution for the real language units is called generation, i. e. you generate real language phrases using the pattern of a model.

Another method of modeling is induction.

Induction is formulation of abstract model schemes on the basis of specific language material. The correctness of obtained model patterns for the language as a whole is also checked by application of the model to real language formations. Models produced by in­duction are called analytical.

For example, if we analyze the following text using traditional syntactic classes:

"The executive summary is intended to provide an overview of the organization and scope of recommendations in this practical guide­line"

we may obtain the following sequence:

S=Art, A, N, V, V, V, Art, N, Prep, Art, N, Conj, N, Prep, N, Prep, A, A, N.

This sequence of classes is an analytical model of the syntactic structure of the above English sentence obtained by induction.

If then we substitute the elements of this model for other real English words in order to prove the correctness of this model, e. g.

"The heavy trucks are used to carry the loads of sand and trunks of trees to this remote site "

we shall initiate the process of generation, i.e. the model will be­come generative.

This brings us to another conspicuous feature of modeling - the above example demonstrates that the synthetic and analytical mod­els are mutually reversible.

Apart from synthetic and analytical models there are two more types of models distinguished in linguistic science.

These are paradigmatic and syntagmatic models.

For example, the word-building patterns (Verb stem + Suffix - tin = Noun, Noun + less = Adj, etc.) are paradigmatic models be­cause they show word-building paradigms, whereas the models showing the structure of language formations in speech (such as A<=N, N<=V and the like) are syntagmatic because they show the relations between the elements of a syntagma.

Paradigmatic models describe the principles of combining the elements into sets in a language, whereas the syntagmatic ones de­scribe the relations between the elements.

For comprehensive language models it is typical to combine syntagmatic and paradigmatic features in one description, i.e. to show what sets of language elements (paradigms) are combined in speech through this or another relation.

The most well known and complete of such combined models is Mel'cuk's "Meaning-Text" Model widely used in machine trans­lation design.

The last and most detailed and precise type of models is an algo­rithm.

An algorithm is a set of instructions or precisely specified opera­tions comprising certain procedure

An algorithm consists of logical operations and commands.

For example, if you are facing the task of selecting the right key for a door lock out of several keys, you will operate according to the folio wing algorithm:

1. Take a key (command).

2. Insert the key into the keyhole (command).

3. If it fits go to 4., else go to 1. and 2. (logical operation).

4. Unlock the door.

The latter task is the most difficult one because the more specific you become about the observables and rules of transformation of a text in one language into a text in another,the harder it appears to become to find and describe both the observables and the rules.

It seems worthwhile to illustrate the concepts of theory, model and algorithm using a simple example of road traffic.

Let us assume that you need to get from town A to town В by car.

In this case the theory will consist of the basic philosophy of trav­elling by car, stating that you need to use roads suitable for cars and not just drive along the bee line connecting A and Д that while driving you need to observe the traffic rules, that you need to have a driver's license and a sufficient supply of gas.

In other words, having studied the above and other principal observables relevant to your planned journey you need to be able to make the final statement of the theory, i. e. to decide whether you can or cannot safely cover the distance between A and В by car.

The model of your trip will be the road map with traffic signs, showing the motorway connecting A and B.

The algorithm, however, will include a set of car driving instruc­tions and that of the traffic rules at each road sign.

Translation may be regarded as a problem, and the process of translating may be represented as a set of instructions (an algo­rithm) intended to solve this problem.

Before discussing this option, however, it is worthwhile to dis­tinguish between the two principal types of problems - well defined and poorly defined ones.

Well defined problems are those about which we can say, at least in principle, that we have solved them. Well defined problems are solved using algorithms.

Now, is translation a well defined problem? Perhaps, even a beginner would say "no", because not only his or her translation is far from being the optimal solution of the problem, but also that of a highly skilled and experienced translator may be less than optimal.

Moreover, some linguists maintain that a translation as such is impossible and what is possible is a more or less successful interpre­tation of the source text.

However, the truth is that.

At a certain level translation is indeed a well defined problem which can be solved using algorithms. This fact is proved by the ex­isting machine translation systems.

We shall refer to this characteristic of translation in our further discussions and now, before we start to discuss theories and models of translation, it seems rational to try to answer a natural question: "What is translation?" And this will be the subject of our next dis­cussion.

1. 4. Translation as an Object of Linguistic Modeling

Usually when people speak about translation or even write about it in special literature they are seldom specific about the meaning. The presumption is quite natural - everybody under­stands the meaning of the word.

For modeling purpose, however, intuitive understanding is not sufficient; what one needs is a definition.

As you know, translation means both a process and a result. And the first question to be answered when defining translation as an object of linguistic studies is: "Which of the two aspects of translation are we interested in?"

The answer may seem evident: when modeling translation we are interested in both its aspects.

First of all, we are interested in the process because it is the process we are going to model.

But at the same time we need the result of translation since alongside with the source the translated text is one of the two sets of observed events we have at our disposal if we intend to model the process.

In order to explain translation we need to compare the original (source) text and the resulting (target) one.

However, the formation of the source and target texts is gov­erned by the rules characteristic of the source and target languages.

Hence the systems of the two languages are also included in our sphere of interest. These systems consist of grammar units and rules, morphological and word-building elements and rules, stylis- tical variations, and lexical distribution patterns.

Moreover, when modeling a language one should never forget that language itself is a formal model of thinking, i. e. of mental con­cepts we use when thinking.

Since a language is a model, linguistic modeling is the modeling of a model

In translation modeling we deal with two languages, i.e. with two formal models, and to verify the information they give us about their modeling objects (concepts) we should model the con­cepts in a different way rather than only through the relevant lan­guages, i. e. through situation, through speakers' image, and their behavior.

Having considered all this, we shall come to understand that as an object of linguistic modeling translation is a complex entity con­sisting of the following interrelated components:

• elements and structures of the source text;

• elements and structures of the target text;

• transformation rules to transform the elements and structures of the source text into those of the target one;

• systems of the languages involved in translation;

• conceptual content and organization of the source text;

• conceptual content and organization of the target text;

• interrelation of the conceptual contents of the source and target texts.

In short, translation is functional interaction of languages and to study this process we should study both the interacting elements and the rules of interaction.

Among interacting elements we must distinguish between the observable ones and those deducible from their observables.

Now, let us ask ourselves: "What are the observables in transla­tion?"

Of course, these are parts of words, words, and word combina­tions of the source text.

Besides, there are parts of words, words, and word combinations of the target language (not of the target text, because when we start translating or, to be more exact, when we begin to build a model of future translation, the target text is yet to be generated).

What else? Nothing? No. We also observe the situation.

When we translate at a technical conference we would never translate "conductor" as кондуктор" or "дирижер" and, on the contrary, when translating the speech of a famous conductor, ad­dressing an audience of musicians we would hardly call him "господня провод или господин проводник". Why? Because we observe the situation.

In a way, but only in a way, you may call observable that part of the background information that refers to situations because you did observe it some time ago and now it is stored in your memory.

All the rest of translation process components are deducible, i.e. the words and combinations of words in the target language which you choose as equivalents and the rules of selection and substitution.

During translation you fulfil the following operations:

• deduce the rules of equivalent selection and substitution on the basis of observed events (source text elements),

• build a model consisting of the target language elements se­lected for substitution and the deduced rules

• and generate the target text on the basis of this model.

In parallel with the generation of the translation model you verify it by applying this model to source and target text elements.

Thus, the process of translation may be represented as consisting of three stages:

1. analysis of the source text and situation,

2. synthesis of the translation model and

3. verification of the model against the source and target

context (semantic, grammatical, stylistic), situation, and back­ground information.

Let us illustrate this process using a simple assumption that you receive for translation one sentence at a time (by the way this as­sumption is a reality of consecutive and even simultaneous transla­tion".

For example, if you received :

"At the first stage the chips are put on the conveyer"

as the source sentence unless you observe or know the situation your model of the target text will be:

"На первом зтапе стружку (щебень) (жареньїй корто- фель) (парезанньїй сирой картофель) (чипсьі) помещают на копвеер ".

Having verified the model against the context provided in the next sentence (verification against semantic context):

"Then they are transferred to the frying oven "

you will obtain

"На первом зтапе парезанньїй сьірой картофель помещают на конвеер ".

Object	Is it usually fried?
стружка	No
чипсы	No
жареньїй картофель	No
щебень	No
нарезанньїй сьірой картофель	Yes

In the case we have just discussed the translation model is veri­fied against the relevance of the concepts corresponding to the word "chips" in all its meanings to the concept of the word "frying".

Let us presume now that we have only one word at a time at the input of a translation automaton (which is indeed so in many ma­chine translation systems}. If the first input word is "blue" the model of translation will include at least two Russian equivalents and their declination paradigms (to save the space we show the paradigms only in part and only for Masculine Gender):

Russian Equivalents and Declination Paradigms
син-ий	голуб-ой
син-эго	голуб-ого
син-ему	голуб-ому
etc.	etc.

Besides, the translation model will include the rule of Russian grammar stating that the adjectives must agree with nouns in Gen­der, Case and Number.

Then, if the second input word is "flower", having verified the model of "blue" against it, we shall obtain the equivalent голубой (синий) цветок which is still ambiguous. We may call this process verification against grammatical context.

Further verification against the semantic context or situation may help to distinguish between "голубой" and "синий".

Verification against semantic and grammatical contexts is per­formed either simultaneously (if the grammatical and semantic ref­erences are available within a syntagma) or the verification against semantic context is delayed until the availability of a relevant se­mantic reference which may be available in one of the following rather than in one and the same sentence:

long stick - длинная палка long run - долгий пробег	grammatical and semantic context in one syntagma
The results are shown in the table Результатьі показаньї в таблице (столе)	grammatical and semantic context in one sentence
The tanks were positioned in specially built shelters and the tank operation proved successful. Танки (емкости, цистерни) бьши помещеньї в специально построенньїе укрьітие и танковая операция прошла успешно.	grammatical and semantic context in different sentences

With these simple examples I want to stress a very important fact in language and translation modeling: the co-occurring words or the words situated close to each other in a text have invisible "pointers" indicating various kinds of grammatical, semantic, and stylistic information.

This information is stored in human memory, but an automaton is completely "ignorant" about it and the principal task of modeling is to visualize this information.