The onomasiological and semasiological perspectives

One of the procedural tasks in lexicography, including terminological lexicogaphy, is to define the structure of lexicographic objects. These objects include:

1. Lemmata
2. Words (including terms)
3. Morphemes, morph variants
4. Word classes (semantic, syntactic, morphological fields/types)
5. Representations:
   1. Phonological
   2. Orthographic
   3. Semantic
   4. Conceptual

The organisation of a lexicon or terminology can have any of these objects as its primary unit. However, traditionally two main structural principles are generally recognised in lexicography; a fortiori they may also be applied to terminology:

1. The semasiological principle. Semasiological organisation is based on the forms of words, that is, either on their orthography (the most usual choice) or on their phonology. In many languages there are complex relations between the orthographic variants of a word, many of which are mediated by the morphology of the language; the same applies to phonological variation. Standard alphabetic dictionaries are the most straightforward example of semasiological organisation; the choice of lemma or headword, and sub-lemmata, is determined on morphological grounds.
2. The onomasiological principle. The grouping of words in terms of their meaning, i.e. on conceptual or semantic grounds, is the second main principle of lexicon organisation. Onomasiological organisation is based on semantic fields or on shared properties of meaning. These semantic fields are hierarchically organised in terms of three basic types of relation:
   1. Taxonomy: the is-a relation of semantic networks, defining elements of sets, and subsets of supersets.
   2. Mereonomy (meronomy): the part-of relation of semantic networks, defining parts of wholes, and sub-aggregates of aggregates.
   3. Predication: the has-prop relation of semantic networks, defining the characteristic property (defining property) of a set.

   Within the onomasiological principle, strictly speaking two further criteria can be distinguished:

   1. Conceptual. Semantic objects are defined as language-independent.
   2. Relational. Semantic objects are defined as language-specific.

   For terminological work, the first assumption, that semantic objects are language-independent, is a useful idealisation; the second assumption is, however, often closer to actual practice and to the complex meanings of words in different languages.

   For multi-lingual terminology, a set of relations based on the frequently cited Vauquois triangle can be defined: at the peak of the triangle (see Figure 1) represents universal concepts; the levels below this represent language-specific notions which need to be connected via explicit transfer relations (transfer rules).

 
Figure 1: Vauquois triangle for terminology 

The triangular model is also useful as a visualisation of the relation between special languages or technical sublanguages as opposed to general everyday language as used in informal speech and writing, and also as a visualisation of the relation between the languages of different disciplines which refer to the same objects in the real world or to the same concepts, but in a different terminological framework. A good example of the second kind of relation is the finite state machine (FSM) of computer science, most generally referred to in contemporary computational linguistics as a finite state transducer (FST); a probabilistic variety of FSM is referred to in pattern recognition end spoken language engineering as a Hidden Markov Model (HMM). The formal basis for formalising the objects referred to by these terms is the same; the differences are differences in the terminological traditions of different disciplines.

It will be assumed here that the same general relation types are used to classify types of transfer relation and to structure concepts. The task of re-using the terminological basis in the Handbook of Standards and Resources for Spoken Language Systems can be defined in terms of this relation; the distance between Handbook usage and systematised terminology is rather small.

In deciding on a terminological model, fundamental decisions on these points have to be made. Further issues to be dealt with at a later stage include:

• a unified sign-based approach;
• terminological morphology;
• terminological semantics;
• terminological acquisition;
• automatic term extraction.