There are phenomena of variation of shapes which are best treated by finite automata (DAGs). We mention a few examples:
- Variations of shapes for a given technical term, that morphological variation and/or abbreviations. The first two graphs given in annex 3 provide examples of this very common situation (technical terms: perfect Penrose tiling system and insuffisance de la glande thyroåde)
Is is important to notice that the variants of a given technical text can be seen as constituting an equivalence class of forms for the term. In principle, the variants are all synonymous. Hence, it is possible to add to the graph paths occupied by synonyms unrelated in form to the variants, which is about what happens with hypothyroidy. These equivalence classes could constitute a basis for constructing new types of thesaurus, well adapted to applications such as information retrieval.
- A common use of local grammars is for semi productive sets of expressions, apparented in shape and meaning. We give the following examples in annex 3:
- Local grammars can also be used for the detection of new terms, especially in the case of compound utterances. The last two graphs of the annex 3 are grammatical graphs that represent shapes of compound nouns built by means of nouns and agreeing adjectives, we give the graph of binary terms and the graph of ternary shapes.