An overview of the MULTILEX architecture

• The MULTILEX model was designed from the start as multilingual. This viewpoint has motivated important choices such as
• flexibility of the model which should account for all the languages of the European community,
• identification of information nodes to allow for language equivalences,
• language-independant overall architecture.
• As a multifunctional model, MULTILEX can integrate information of a variety of theories and information intended to many applications (text analysis, automatic or computer-aided translation, etc.)
• The MULTILEX model is described as an SGML DTD (the MLEXd) for reasons of portability. Its design is independent from any internal database representation.

Global structure of MULTILEX dictionary A monolingual dictionary designed in accordance with the MULTILEX standard consists of:

Language

(obligatory)
The name of the language which the entries of the dictionary belong to.

Language-specific information

(optional)
The core of the standard is language independant. However, some features are not. Specific attributes have to be defined or inhibited for each language. For example, gender is not relevant for nouns nor adjectives in English but it is in Italian, French, Spanish, German, etc. This is also true for values. The values for gender in German are masculine, feminine or neuter while in French, Italian and Spanish, neuter is irrelevant.

Standardisation does not mean a frozen model. Each dictionary designer is free to define his/her sub-model with specific constraints providing that it complies with the standard. In this section it is possible to define local specific features:

• add values to the lists which are declared open,
• inhibate values in the lists declared free,
• add an attribute or sets of attributes for unforseen or insufficiently structured areas,
• change, when permitted, the status of an attribute or sets of attributes this status can be:
  • obligatory
  • inhibited
  • default = value
  • optional

Rule specifications

(optional)
Rule specification which makes it possible to annotate the dictionary with grammatical information. It is often necessary to refer to a rule which is applicable to an item without having to describe this rule. A reference name is generally used. For example, plural formation rules imply a series of manipulations on the stem and the endings which are performed by the client NL application. An identifier of the rule is preferable to the actual name of the software module used in a specific application if the lexical resource is to be re-usable. An even better identification is to associate to this identifier the description of what the rule does in a simple and formalised way. This increases the readability of the data and its portability because it is possible to change the coding system for the rules without loosing functional description and without having to change the value names in the body of the dictionary.

Body

The body of the dictionary consists of the entries, i.e., the objects allowed in the dictionary by the standard.

 
• A two nodes model
• Linguistic levels
• GPMU and CGPMU
• The LU node
• Cross-references
• Terminology
• The multilingual architecture
• Conclusion