Skalmowski - 2

For this paper it seems advisable to hold apart two notions: that of the "concept and that of the "semantic unit". Given a generative descriptional device G /grammar/ and a pro­jective system of the type proposed by Katz and Fodor S /seman­tics/ we can describe a semantic concept in a'language L as a set of n-tuples of symbols from G and 3, ordered or par­tially ordered by the relations which define the formal rules of these systems, and having a common derivation in 2. This broad frame allows us to regard as a concept every dictionary entry -except for-the "grammatical words" which do not possess any de­rivations in G - and leaves us a wide margin of freedeom in constructing arbitraty "conoe: t -systems- with a priori estab­lished features.

In a similar way we may describe a semantic unit as a set of n-tuples of G-symbols, G-rules of word formation and g-symbols, ordered or partially ordered by means of rela­tions which define the formal rules of these systems, and having a common derivation in G from some 3—symbol uniquely related to some ü-symbol. This allows us to relate with tue notion of a semantic unit the linguistic notions of morpheme /or mere strictly: semanteme/ and of "word family", defined in terms of grammatical derivations.

The thesauric approach to the problem of meaning . in i:T /s.e.g.3/ pays tribut to the idea of ordering the symbols within the concepts, but at the same time it brings to light the problem of multiple meaning. This problem has been much discussed already /s.e.g.4/, but it is still far from being solved in all its aspects. Generally speaking the main difficul-

Skalmowski - 3 ty arises from the fact that the "concept-systems" of languages are not isomorphic and even if we manage to bring them closer together there remains some amount of "looseness" within the concepts themselves, giving rise to the problem of "true" mul­tiple meaning. The "contextual'' multiple meaning may be resolved' - in principle, at least - by extending the notion of concepts both 'in the source and in the target languages to whole sentences or even larger utterances; this is allowed by our " broad' treatment of this notion, not specifying the maximal size of the n-tuples of symbols. By this extension the inner structure of concepts makes the relations defining the isomorphism of the "concept -systems" more apparent; thus even such cases as the adequate translation of the Russian U$JteneHU9&s the English "changing /the order of integration/"and "varying /argument/" are theoretically resolvable. Yet there exist instances where the extension of concept would have to go beyond limits and to involve the whole language: these are cases of "stylistic' difference in which there are not apparent reasons for choosing one of the possible synonyms instead of the other but where tue difference is distinctly felt by competent bilingual speakers. The problem is important for the translation of literary pieces, especially poetry; by the present stand of it is still an "academic" problem, of course, but it exists after all. It may­be best illustrated by the question whether tuera are "betterand "worse" translations of nonsensical expressions, such as the famous "furiously sleeping ideas", „egative answer would mean that every translation is equally good, which in turn would mean that only "meaningful" sentences are translatable; in that case

Skalmowski the ivlï problems would be "enriched" with the whole load of phi­losophical questions - an embarassing development, certainly.

Vaguely felt differences between the intrinsic "semantic values" of different elements of language have given rise to the notions of "size" or "content" of semantic elements /5/ and several attempts - both to define these notions and to furnish models of the underlying mechanism have been made /5,6/. The main assumption - based on observations of ,/illis - was that there existed a "natural hierarchy" of concepts in natural languages,' forming a tree or at least a lattice with some de­finite statistical properties.

The present paper gives some results of an in­vestigation undertaken in order to test this hypotheses. Because of the marvelous clarity of the grammatical structure Arabic has been chosen as a "laboratory example". ^bout 90,j of Arabic semantemes are verbal roots, with very few exceptions consisting of three consonants G^-G^-C^; the usual dictionary form is the 3 per s. sg. maoc. nerf, of the form C.aC aC-.a, s.<-.

kasara "to break" /lit. "he has broken"/. There are more than ten different verbal stem-patterns i.e. word formation rules, raodyfying the basic meaning of the root in a specific way; thus the stem-pattern II: C ,aC0C..aC-,a adds to tue basic meaning the shade of intensity, e.g. kasara "to break' kassora "to smash"; the stem-pattern III is conative, the IV - causative, etc.

. nil the trilitersl verbal roots in the ^rabic vocabulary have been divided into se para to classes according to their ability to form s = 1,2,..., n different stems. ..e only

Skalmowski the number of stem-patterns was considered and further applicabl v/ord formation rules /substantivisations, adjectivisations etc./ were disregarded this classification is a very rough approxi­mation to the hypothetical underlying hierarchy. It has been assumed that the number of stem-patterns defining a given class may be approximately viewed as an exponent of the "content" or "semantic value" of the semantic units belonging'to this class and that - if the hypothetical hierarchy was really based on this principle - the number of roots with greater s should be smaller than' that with smaller s. Baranov's Arabic-Russian Dictionary /7/ has been used for counting the roots and it has been found that the relation between s /the number of steui--patterns characterizing the given class/ and r /the number of roots belonging to this class/ was not only inversly proportiona but. also nearly functional and that the distribution of roots in the Arabic vocabulary may be described as a simple function r/s/ = N/As'" +Bs +C/, where Ii is the sum-total of roots and A, B and G are specific constants. The goodness of fit has been tested by the chi-sapuare distribution and it has been found that the differences between the empirical data and the theoretical distribution - except for one value - do not exceed 0.3 signifia ance .level.

In order to estimate the possible differences between particular dictionaries - which could arise from differences between the materials used for their compilation -two samples of ca. 700 items each have been taken from two different dictionaries /7,Q/ and the distribution of roots in them compared with each other and with the over-all distribution

Skalmowski - 6

All the distributions show a striking similarity, rendering nearly identical chi-square values. '

This result is a strong argument for the general validity of the discussed distribution in Arabic - and this fact in its turn speaks in favour of the existence of "natural hierarchies" of the semantic units in general.

It seems very probable that similar regular dis­tributions might be found in other languages, too - perhaps the ensemble of the "semantic parameters" would have to be much wider and the "trial and error" investigations would require more time but the whole work can be easily mechanised. The idea of interconnections between the syntactic and semantic structures of language is not new in structural linguistics /s.9 and 10/ and investigations along these lines have already been led in the domain of computational linguistics under direction of P.Garvin /11/« % suggestions go towards discovering such regular

Skalmowski - 7 distributions which would facilitate the task of finding more strict correlations between the synonyms within particular con­cepts on computational basis. The underlying assumption is that the "universes of discours" in various languages are of about the same "size" /whatever it would mean - but such an assumption is tacitely made in every translation/, and that the semantic units underlying the components of concepts are ordered according to their "content", so that the problem of "true" multiple meaning in certain cases may be solved by means of matching the components of concepts of the source and target languages on the basis of their "semantic value".

As an illustration let us consider a few equivalent English verbs in two different translations /A. -12, . -13/ of the Koranic Sura 84, being translations of Arabic verba derived from roots all belonging to the same class /5 stem -patterns/, i.e. according tc our assumption having about the same "semantic value". The "value" of corresponding ünglish verbs has been tentatively estimated by the number of different sub-entries in Chambers's ^k^ Century Dictionary /numbers in brackets/: Arabic English infatara

Skalmowski - 8

The applied "method" being unsystematic and ad hoc the example allows no generalisations but it may illustrate our argument that the problem of '"true" multiple meaning arises in cases of "expressive language" from the fact that even when the concepts of source and target languages agree there is no correlation between their respective components except for differences between their "value", based on differences on tue paradigmatic level. Thus e.g. for the concept "applying heat on something" two different semantic units could have been arbitrarily chosen by the two interpreters, as they regarded th subsets of synonyms within the concepts as unordered, My suggestion is that these subsets might be at least partially ordered by means of the intrinsic value of the semantic units underlying them and that correlations between them might be established in more objective terms of numeric measures of their content.