Preliminary Recommendations

 

The proposal phase

The common set of specifications, resulting from an analysis and evaluation of the results of the comparison, and constituting the EAGLES proposal, is contained in the bottom box of the synoptic tables (see table 2), a set of rows named EAG-L0, EAG-L1, EAG-L2a and EAG-L2b in the second zone of the tables. The features are indeed articulated on different levels, corresponding to different degrees of obligatoriness. As already pointed out (see Leech (to appear), Monachini & Östling (1992b) and Leech & Wilson (1993b)), different levels of constraints can be isolated in the morphosyntactic encoding of a category and, therefore, different levels of standardisation can be suggested.

 

Description of the second zone of the tables

In the EAGLES proposal, the specifications appear as follows (see table 2):

• Level 0 (EAG-L0) contains only the category (or PoS), i.e. Noun, Verb, Adj, etc., presented also in the document by Leech & Wilson (1993b) as the only obligatory features;
• Level 1 (EAG-L1) presents the grammatical features, i.e. the agreement features such as Gender, Number, Person, etc., which are usually encoded in lexicons and corpora: these are considered as recommended features (in Leech & Wilson these are defined as optional) constituting the minimal common core set of features for the PoS;
• Leech & Wilson `Special Extensions' are presented here on two levels; they are in fact exemplary of current practices from two different points of view:
  • Level 2a (EAG-L2a) contains information which is pertinent to all or many languages and which is considered useful and easy to standardise, but is either not yet usually encoded by current practices or not purely morphosyntactic (e.g. countability for nouns): these are to be considered as optional features. As a rule of thumb, a value is put here if it is relevant to more than three languages.
  • Level 2b (EAG-L2b) presents language-specific features, not belonging to the set of recognised common features. The values presented in this row are labelled with the language they are pertinent to.

The names of the attributes and values are presented in the tables in an abbreviated form, due to the limits allowed by the page: most often, the abbreviations are self-explanatory (e.g. Gend stands for Gender and Numb for Number). The full names of the attributes with the relevant values are listed in the Comment section which follows each of the tables. The abbreviations of those attributes that do not appear clear from the tables can hence be deduced from there.

As they are intended now, the abbreviations do not represent in itself a proposal of how to encode concretely these attributes in lexicons, i.e. they are only labels to represent a distinction that in the various European concrete practices can be marked using different symbols. For example, for the singular value of Number, any of singular, sing, sg, s or 1 can be used. What is important is the mappability from a concrete system of labels to the set of labels proposed here. This can serve, therefore, as an interface or interchange proposal.

Having a multilayered or hierarchical structure, instead of a flat one, gives more flexibility to the proposal and allows the user to choose the most appropriate level of encoding (which may vary, e.g. according to different applications). The idea is that, going from Level 0 to Level 2, the amount of information is increased and more granularity is achieved. The use of all three levels allows the representation of more information, whereas a description at Level 1 is a subset of a description resulting from the application of all three levels. Moreover, a multilevel proposal provides an easy framework for specific extensions and updating, as well as permitting a comparison of different schemes (at the lowest common level).