The expanded system of the coordinated dictionaries and its using for universal semantic coding and translation of the polysemous source text

By Panich Iuli,
Russia

iuli-panich [at] yandex . ru

Get the List of 4,500+ Translation Agencies Now!
No Recurring Membership Fees!

Universal semantic coding

Universal semantic coding (continuation)

Universal semantic coding (Appendices)

Читайте также версию на русском языке

Annotation. Semantic coding is based on supplementing the polysemantic words and word-combinations by components of dictionary entries that the author chooses in the basic explanatory dictionary of the source language (the native language for the author). The universality consists in subsequent machine translation into other languages using bilingual dictionaries coordinated with a basic explanatory dictionary. The process of coding takes into account not only the wide range of meanings of polysemantic words, but also the ambiguity or incompleteness in the expression of actions and states, where in one language such features as gender and number, character of action and state, etc., are determined only by the context, while in another language they may be reflected in concrete grammatical forms.

Contents:
The presuppositions.
Essence of the proposed method. A preliminary identification of the source text.
Characteristics of the expanded system of the coordinated dictionaries.
The special sections of differentiated (morphological and syntactic) meanings in the coordinated dictionary entries.
The semantic coding of the source text.
(continuation)
The translation of the encoded source text into target languages.
Some characteristic examples explaining operations of semantic coding and translating
Some considerations of general character.
(Appendices)
(examples of coding and translation, fragments of coordinated dictionary entries).
App. 1. The phrases containing polysemantic words.
App. 2. The phrases containing polysemantic set expressions and set expressions characterized by the structural formula with variable lexical components.
App. 3. The coding and translation with using the differentiated (morphological) meanings of the verbal forms.
App. 4. The coding and translation with using the differentiated (morphological) meanings of the pronouns.

* * * * *

The presuppositions. As a rule, polysemantic words and expressions inevitably exist in source texts. Translation of these words and expressions depends on the context. Analysis of the sense and form of a wide context is impractical with traditional machine translation programs.

Many source texts published in the world require translation into other languages. Among these are articles in national scientific magazines, descriptions of inventions published in the regional funds of patents etc. All are of interest for users speaking other languages.

Therefore, the source text which can be potentially translated into many target languages should contain certain

additions which, in a universal form, would indicate polysemantic words’ and expressions’ actual semantic meanings.

Such additions should also reflect the grammatical (morphological and syntactic) features of the wider context. During translating into target languages these additions allow to choose the corresponding grammatical forms of words or words' groups and also to choose syntactic links between words.

Such semantic additions eliminate both the polysemy of the source language, and the mutual ambiguity arising during translation; it will allow to receive correct over the sense and over the form the machine translation into any target language.

Some rather abstract methods of semantic coding of source text are known [1-3, 3*].

One of the early methods, based on semantic analysis of polysemantic words and their semantic coding, is described in US Patent 5,285,386 “Machine translation apparatus having means for translating polysemous words using dominated codes”.

This method of semantic analysis is universal in that it does not depend on the target language of the translation.

Some codes of "philosophical" nature are offered. This method uses four classification categories to present all information on a word: the great category (it is designated by the first numeral), middle, small and thin categories of classifications.

All words in a thesaurus are classified in ten great classifications, including "character / nature", "property", "change", "action", "feeling", "person", "propensity", "society", "culture", "subject".

Each great classification is divided into ten middle classifications. In this embodiment a code with a symbol S is given by the following form:

S0 belongs to "character/nature"

S02 belongs to "weather" in "nature"

S028 belongs to "wind" in "weather"

S028a belongs to "pressure" in "wind" ….

The given codes of classification have hierarchical character: the range of meanings for the upper semantic code is wider than for the lower semantic code.

As has been noted, the method described in this patent can have a universal character; however the offered hierarchical system of the codes describing semantic meanings of words and dependences between words has too abstract character, i.e. it is far from the real problems of translation of expressions and word-combinations with their various usage and complex syntactic structures, etc.

Elements of semantic coding of similar abstract character are offered in US Patent 5,523,946 “Compact encoding of multi-lingual translation dictionaries”.

Each word is a file containing information codes pertinent to some base abstract language; this information is based on the concepts of type - part of speech, gender, number, punctuation mark, area of meaning of a given word (for example, a field of knowledge, actions, …).

For example, the code for the verb “to know” may include the attribute (sth) which limits the field of knowledge.

The coding is done by an editor of the source text.

The following source phrase is given as an example: "to have down pat, be an expert"; this phrase is divided into the following symbols: "to" "have" "down" "pat" "," "be" "an" "expert".

The codes for each symbol and its concept have the following form:

1991 "to" 872 "have"  501 "down" 18005 "pat" 17 ","

427 "be" 95 "an" 2407 "expert".

The string, representing this phrase, contains the following numeric sequence: 1991 872 501 18005 17 427 95 2407.

As stated in the description, the given translation file consists of about 6700 groups of concepts.

In sum, this method contains elements of semantic coding in the form of concepts of an abstract basic language.

This coding, however, is also very superficial inasmuch as it is not concerned the concrete semantic meanings of words, but is based on general attributes (part of speech, gender, field of knowledge, etc.).

Finally, there is the Universal Networking Language project (UNL) [3,3*], which has being in development for a long time. This project aims to create a universal artificial language by developing codes, which, in some abstract form, would reflect the set of semantic and grammatical features of a source language, and, at the same time, reflect translational specifics of a set of target languages.

In other words, these artificial semantic codes would have to be refined and coordinated as much as possible, i.e. they would need to simultaneously reflect the intra-language polysemy and the mutual ambiguity, arising in translation[16, 16 *].

A special apparatus of coding of semantic meanings is elaborated for this purpose. The UNL has its own dictionary UW (universal word) made up on the basis of English, which is an input. Specifically, it is English words that are used to determine semantic and syntactic constraints or restrictions.

The text on UNL is universal words with additional symbols of type:

icl (inclusion), 

aoj a thing which is in a state or has an attribute: aoj(red (aoj >thing), ball(icl > thing))… “ball is red”,

… etc.

Each meaning can have also a set of particular meanings, for example:

seq (occur, occur) / seq (occur, do) / seq (do, occur) / seq (do, do) / seq (occur, (aoj>thing))...

Example of designation:

“because Mary arrived, John is happy” … agt: 01 (arrive (icl > occur), Mary (icl > person) agt: 02 (happy (icl > do), John (icl > person) rsn (:02, :01);

“horse” … horse (icl > animal) horse (icl >male)  horse (icl > apparatus).

Codes of attributes are added to a universal word and they specify variants of use:

category of tense @future,… /  modality @obligation, … / character of expression @emphasis, @focus, …, @begin-soon, @begin-just… @progress… @complete, @state, @repeat … @generic, @indef … @affirmative, @confirmation, @exclamation, … etc.

The sentence “The dog caught a wild cat” will have following UNL-representation:

[S] agt(catch(icl>#event).@past.@pred.@entry, dog(icl>animal).@def)
obj(catch(icl>#event).@past.@pred.@entry, cat(icl>animal).@indef)
mod(cat(icl>animal).@indef,wild(icl>#state, ant>domestic)) [/S]

These examples indicate that artificial language has the special dictionary structure, special forms of representation of semantic meanings and also has own morphology and syntax; all these features are expressed as complex chains of restrictions and attributes.

The UNL is based on double machine translation (see the scheme): (I) from the source language into an artificial language (Enconventer) and (II) from an artificial language into target language (Deconventer). On the first stage the author can make the interactive changes directly in the text of an artificial language by means of operations of the direct translations and the reversed translations (Author / Editor).

It is possible to point out the following properties of system UNL:

a) as already noted above, the analysis of sense and the form of a wide context is practically impracticable by means of programs of machine translation.

The double machine translation stipulated in system UNL will lead to inevitable summation of mistakes in transfer of sense of the text and to mutual influence of ambiguities, arising on each of stages; in other words, the mistakes arising at the first stage, can generate a circuit of mistakes at the second stage;

b) the author has an possibility to make some changes in the text in an artificial language as it is stipulated in system UNL; but for this purpose the author should know professionally this language, i.e. should be able to indicate the real deviations which have arisen between the source text and its reversed translation by means of combinations of symbols;

с) it is known, the operations of direct and reversed translations are carried out a phrase by a phrase and they demand a multiple recurrence; these operations can be applied only at translations of short texts and may have extremely unstable character (similarly to a control system containing elements with unpredictable reactions). In others words, if the basic programs of the machine translation (Enconventer, Deconventer) make the distortions of sense of the text, then the essential increase of a level of accuracy due to the interactive sanction (Author / Editor) will not occur;

Probably, the universal language UNL can be used for transfer of the sufficiently short and "monosemantic, unequivocal" texts.

To sum up: one can assume that the abstract methods based on artificial codes allow to solve a problem of universality, but they do not contain the elements allowing to raise accuracy of machine translation.

Essence of the proposed method. A preliminary identification of the source text.

Proceeding from the general reasons, a source polysemy (or a source ambiguity) leading to wrong interpretations of words and word-combinations should be eliminated prior to the beginning of machine translation.

The information necessary for a very precise semantic coding of the source texts is already present in the existing explanatory dictionaries and bilingual dictionaries. The detailed presentation of semantic meanings in an explanatory dictionary characterizes the internal polysemy of the source language. The illustrative examples ... below we use ill.e.(s), included in the dictionaries, are important in that they reflect the various forms of usage of word and word-combinations; the translations of the specified examples in the bilingual dictionaries reflect the mutual ambiguities of the source and target languages.

The basic idea is to use the elements of the explanatory dictionary of the source language instead of some artificial codes; these elements can naturally reflect the sense of a wide context.

An author would need to carefully examine the meaning of a given word in its context, compare that to the available dictionary entries, and, then, key in the actual meaning of the word, or a corresponding illustrative example, as a supplement to the word, thus insuring a proper translation into target languages.

In other words, this is a preliminary identification of semantic meanings, forms of use, etc., by means of an explanatory dictionary of the source language.

The question is how to transform the various elements of dictionary entries, including the aforementioned ill.e.(s), into universal semantic codes. In other words, the problem is how to coordinate these elements with the translations contained in the bilingual dictionaries. To resolve this problem we propose a system consisting of a basic explanatory dictionary of the source, native language, which we shall call further a Coordinated Explanatory Dictionary (c.e.d.), and a set of special dictionaries of source language - target languages, which we shall call further the Coordinated Bilingual Dictionaries (c.b.d.) [19, 19*].

Characteristics of the expanded system of the coordinated dictionaries. 

The following elements of each c.b.d. shall exactly match the c.e.d.:

- the vocabulary of source language, a sequence of the vocabulary entries,

- the sequences of semantic meanings, each identified by the same digital, alphabetic and symbolic designations; this also concerns the particular and closely related sub-meanings of each semantic meaning,
- the ill.e.(s) reflecting different word usage at concrete semantic meaning,

- all expressions, their semantic meanings and their own illustrative examples,

- all grammar/usage commentaries on the headword, derivative words, compound words, etc.

The specialized and thematic sections of the dictionaries shall similarly match.

For illustration let us present some fragments of the coordinated dictionary entries.

The c.e.d. thus combines the function of an explanatory dictionary with those of a translation dictionary. That is, the c.e.d. must reflect those elements of the source language, which have special meaning at least in one of the target languages belonging the system of coordinated dictionaries of this source language. Really, if ambiguities of one type do not depend on target language, then the ambiguities of other type will arise only in translating into the certain language.

Thus, together with the multitude of semantic and grammatical features of the source language, the c.e.d. should reflect the translation - specific features of all target languages.

Therefore, the entries should be “maximally differentiated”, i.e. they should simultaneously reflect both the intra-language polysemy and the mutual ambiguity, namely:

- the c.e.d. entries should contain the non-phraseological expressions having the especial translation (the non-word-for-word translation) at least into one of the target languages out;

- the c.e.d. entries should reflect the meanings of morphological nature, due to the presence in any of the target languages of separate grammatical forms, which are absent in the source language;

- the c.e.d. entries should also reflect the meanings of syntactic nature, for example those indicating possible syntactical links, clarifying syntactical ambiguities in the source language, etc. With this purpose, the differentiated (morphological, syntactic) meanings are entered into dictionary entries in the form of the special (additional) section that follows those semantic word meanings to which this section is general. These sections have a symbolic designation (morph_m) (an abbreviation of morphological meanings) and (syntax_m) (from words syntax, meanings).

The following example reflects a polysemy of pronoun you and also reflects the coordination of entries when in one of the languages semantic word meanings are defined by a context, however in other languages these meanings have the separate grammatical forms:

In the given example the pronoun you replaces the person or group of persons of masculine gender or feminine gender (what depends on sense of the text). In translation into Russian a pronoun you correspond two forms - ты, вы. At last, in translation into a Hebrew  a pronoun you corresponds to four forms expressing a masculine gender and a feminine gender both in a singular and in plural:אַתֶן  / אַתֶם/ אַת/ אַתָה. 

How we see, the specified features of target languages are reflected in English c.e.d. where the meaning of a pronoun you is presented in the form of four private meanings (a)…(d) - it takes into account the specified features of two target languages – Russian and Hebrew.

An expanded fragment of coordinated entries showing the differentiated nature of dictionary entries is presented below:

As follows from the table, the sections of illustrative examples and of the non-phraseological word-combinations are coordinated and as much as possible expanded; the use of elements of these sections at semantic coding allows to take into account the features of the use of words.
Some explanations, concerning the section of differentiated (morphological, syntactic) meanings, will be presented below.

The structure of systems of coordinated dictionaries is presented on Fig.1.
As it has noted been above, the entries should be “maximally differentiated”, i.e. they should simultaneously reflect both the intra-language polysemy and the mutual ambiguity.

The account of the all types of the ambiguities somewhat expands volume of dictionary entries. In this connection a system of coordinated dictionaries of given source language may consist of several subsystems of the coordinated dictionaries - for related groups of target languages (Fig. 2).

In the some cases, the subsystem can include only two languages - source and target; it means, that the bilingual dictionary, only one in this subsystem, should be coordinated with an explanatory dictionary.
The coordinated dictionary entries are in details described below.
Let's explain in more detail the sense of mentioned special (additional) sectionсs.

1. The differentiated (morphological) meanings.

We are concerned here with words, which can express, depending on context, different grammatical attributes (i.e. the meanings having grammatical nature), but they have no corresponding grammatical forms in the source language. However in the some target languages there are the separate grammatical forms of a word and / or connected words for the expressing these meanings.

These groups of words may include the following:

1) personal and reflexive pronouns, which, in a given source language, have no separate forms to denote the person of masculine or feminine gender / the person or a group of persons / a group of persons of masculine or feminine gender / the person or an inanimate object, and also possessive pronouns, which, in a given source language, have no separate forms to indicate the gender and / or grammatical number of the belonging object, or the separate forms specifying a gender of the person to which it belongs (i.e. when the meaning depends on who or what is being described in context); at the same time those or other target languages can have the pronouns (and / or words, contextually associated with them) which have the concrete grammatical forms to express above mentioned attributes.

Let’s illustrate this with other characteristic example of the possessive pronoun-noun yours. This pronoun does not have any other grammatical forms, but in a combination with a context may express:

- masculine / feminine / neuter gender of belonging object or plural of this object; this must be taken into account when translating into Russian, where one of next pronouns may correspond, depending on a context, to the pronoun yours твой, твоя, твоё, твои …;

- a person or a group of persons to whom the belonging is expressed; this, too, must be taken into account when translating into Russian, where one of next pronouns may correspond, depending on a context, to the pronoun yours твой, твоя, твоё, твои; ваш, ваша, ваше, ваши; 

- masculine / feminine gender of person or persons to whom the belonging is expressed; this must be taken into account when translating into Hebrew, where one of next pronouns may correspond, depending on a context, to the pronoun yours  שֶלךָ - belongs to a male, שֶלָך - belongs to a female, שֶלָכֶם - belongs to a males group, שֶלָכֶן -belongs to a females group.

The described polysemy of the pronoun yours, having a grammatical nature, is reflected in the following dictionary entry:

As we see, in Hebrew the dependence of personal pronouns on the gender and number of the subject is even greater than in Russian or English.

2) verbs, participles, verbal adverbs, which in the source language have no separate forms unambiguously expressing the character of an action and /or a state, reached as a result of the action (i.e. it is possible to judge the character of the action and / or the state only from context);

at the same time, the specific grammatical forms of some verbs, participles and verbal adverbs are used in some target languages to express the shades of the meanings of the actions and / or the states required by the source text;

below are some characteristic examples of grammatical meanings in English and Russian:

a) in Russian a verb, in context, may express one of the following meanings, which correspond to a specific grammatical form in English ([12], p. 113-162):

- (неопр.) … (the separate group of the Indefinite Tenses is used for this purpose in English);

- (длит.)… (in English this is expressed by the separate group of Continuous Tenses);

- (к опред. м-ту) … (in English this is expressed by the separate group of Perfect Tenses);

- (изв. дл-ти)… (in English this is expressed by the separate group of Perfect Continuous Tenses);

- (состояние), (действие, неопред.), (действие, длит.), (действие, к опред. м-ту)… a Perfective verb in the verbal form of the Passive Voice may express an action or a state, achieved as a result of the action (in English this is expressed by separate group of Passive Voice Tenses, [12], p.166-184);

- (в наст.-будущ.), (в прошл.)… a verb expresses an unlikely or improbable action in a conditional sentence in a Subjunctive Mood with the particle “бы”; the time of fulfillment of the condition depends on the context (and possibly the time is pointed in a sentence previous to or subsequent to the present one).

Also in Russian, a participle (in context) may express one of following meanings to which the separate grammatical forms conform in English ([12], p.113-162):

- (одновр.), (ранее) … a participial form in the Active Voice expresses simultaneous actions or actions preceding the action expressed by the predicate;

- (реальн.)… a participial form in the Passive Voice of the Imperfective indicates the real action expressed by the participle;

- (вообще) … the same as above, but the action happening irrespective of time.

The following table shows differentiated (morphological) meanings in a coordinated dictionary entry of the verb заканчивать:

b) in English a verb, in context, may express one of the following meanings to which at Russian the separate grammatical forms conform ([12]):

- (imperf-ve), (perf-ve) … a verb in the Infinitive or Indefinite Tense form, not having special grammatical forms of the imperfective or the perfective, expresses, in context, an action of imperfective or of perfective; this should be considered during translation, for example, into Russian, in which verbs have separate imperfective and perfective forms;

- (action) the fact or process of doing, (state) the existing condition in which an object is … a phrase in the Passive Voice expresses an action performed on an object or a state of an object, except when some other meaning characterizing an action is not indicated, for example, (imperf-ve)  or (perf-ve);

- (indicative), (subjunctive) … the verb in the form of Indefinite Infinitive or Perfect Infinitive in combination with the verb "should" expresses an action in the form of Indicative or Subjunctive Mood.

Also in English the participle (in a combination with a context) may express one of the following meanings to which in Russian the separate grammatical forms conform ([12].p. 279-287):

- (simult.) … the participial form expresses an action, which is simultaneous with the action of a predicate;

- (preced.) … the participial form expresses an action, which precedes the action of a predicate;

- (imperf-ve), (perf-ve) … the participial form expresses an action of Imperfective or of Perfective.

Examples of differentiated (morphological) meanings are presented below in a coordinated dictionary entry of the verb to ask:

3) the some nouns which in the source language have no forms of a singular and a plural, while in translation to those or other target languages they acquire, depending on a context, separate forms of a singular and a plural.

In the all cases, mentioned above, the author should carefully examine the context so as to add to the word the actual differentiated (morphological) meaning - it will provide a true translation into target languages. 

2. The differentiated (syntactic) meanings.

The meanings which express possible syntactic links of a given word with other words in a sentence are also presented in the dictionary entries.

The specified syntactic links characterize some of the predicative relations (regarding a predicate) and non-predicative syntactic relations (attributive, objective, adverbial).

Use of the differentiated (syntactic) meanings in semantic coding allows to exclude syntactic errors in machine translation of such source sentences that have syntactic ambiguity or syntactical incompleteness; in particular, such source sentences that contain remote or isolated explanatory appendices (attributive, adverbial), which, in turn, contain their own additions.