Main differences

In this section, the main differences between GENELEX and MULTILEX are analysed in the perspective of a synthesis.
The two models have both adopted the traditional linguistic levels, but they differ in the way they organize them. These differences concern the distribution of information around two or three nodes, and also the cardinality of the relations which rules the functionning of the structure.
Although GPMU and UM seem to be rather similar in their contents and their roles, the cardinality of the links between the node (UM or GPMU) has an important influence on the way the relation between the UM or GPMU on the one hand and syntax on the other hand is dealt with. Namely, while each GPMU identifies sets of unconflicting, homogeneous features and constitutes an autonomous entity, a UM contain variant features needing explicit disambiguating links through the linguistic levels.
The main consequence is the close relationship of UM and syntactic information in GENELEX on the one hand and the autonomy of syntax and GPMU - at least in the great majority of cases - in MULTILEX on the other hand. The difference between these approaches also explains the choice of a three-levels GENELEX or a two-nodes MULTILEX architecture.

Syntax

In GENELEX, syntactic information is a layer of the model. A node UM can be attached to zero, one or more nodes USYN. A node USYN is attached to exactly one node UM and to zero, one or more nodes USEM, which is roughly represented:

UM <--0,n-----1,1--> USYN <--1,n-----1,n--> USEM

Syntactic information refers unambiguously to morphology. In MULTILEX, a GPMU can be related to zero (ex: "bacille de Koch" in which "Koch" doesn't need an LU), one, or several LUs. Syntactic behaviours that correspond to nodes LU are described through ``syntactic blocks".

GPMU <--0,n-----1,n--> LU <--1,n-----1,n--> Syntax

Morphology/syntax interaction is therefore described in different manners in MULTILEX and GENELEX. It is necessary to check the possibilities to express the modalities of this interaction within the two models.

Accounting for multiwords

1. MULTILEX and SITE/EUROLANG have jointly elaborated the model of the complex GPMU, CGPMU. It contains:
   • the orthographic description of a multiword unit (= a multiword string);
   • the phonetic description of a multiword unit (= a phonetic transcription of the multiword string);
   • pointers to morphological information about those parts of the CGPMU which have themselves GPMUs or CGPMUs in the MULTILEX dictionary;
   • constraints on the elements of the multiword string (no concrete example available, intended to cover e.g. possibilities of modification, insertion, etc.);
   • a relation with an LU.

   This has the following implications:

   • A multiword unit has its own LU and its own CGPMU.
   • At GPMU level, pointers may be used relating the CGPMU with individual GPMUs; so, morphological information need not be spilt out more than once. .
2. GENELEX has defined models for describing multiword units both at the morphological level and at the syntactic level. Guidelines for selecting the appropriate level where a multiword item is to be coded are provided.

Conceptual description and terminology

EAGLES has still not started the work group on terminology. But it is recognized that the architecture should pay attention to the topic of terminology. See minutes of meeting on 16/4/93.

MULTILEX is experimenting with the introduction of ``specialized LUs (SLUs)'', for the conceptual level. An SLU would be related with several LUs; the LUs related with the same SLU would be quasi-synonymous, and SLU would be a domain-dependent generalization over the meanings of the LUs. SLUs can be cross-referenced.

The same device has been proposed in GENELEX, under the name of CONCEPT.

This descriptive device may be considered as a starting point for work on terminological modeling.