Subsections

• Built-in Semantic Lexicon Database
• Format of a On-line Database
• Format of a Linking Table
  • Access to Linking Tables
• Adding Further Lexicon Entries

On-line Lexicon Databases

 

Built-in Semantic Lexicon Database

The ADT package contains the files vitSemLex.pl, vitValues.pl and vitLinking.pl which define on-line lexicon databases and syntax-semantic linking tables. Currently, there exist lexicons and linking information for German based on [Kasper (2000)], for English based on [Flickinger (2000)] and for Japanese based on [Shimada (2000)]. All three are based on the semantic class descriptions of [Kasper et al. (2000)].

The data which is part of the current ADT package was extracted of the named resources using UNIX shell scripts such as awk, sed, etc. The result is mainly a reduction to relation names and semantic classes. Hence, an on-line lexicon does not contain the whole information of the original database, e.g. decomposition classes are no longer covered. Contextual constraints such as grouping, subordination, etc. are also not part of this database. In general, the values of these constraints are the result of a semantic construction process.

Format of a On-line Database

    An entry in a lexicon database has the following form:

Language:vitSemLex(LabelledCondition,
                   SemanticClass,
                   Class,
                   ConditonCheckingCode,
                   ContextOfCondition,
                   ContextCheckingCode,
                   SortalRestrictions) :- Goals.

As can be seen, Language defines a module in which a lexicon entry vitSemLex/7 can be accessed.

The first argument is always a labelled condition LabelledCondition which should belong to the defined language.

The second argument contains the semantic class SemanticClass of the condition. The potential values are described in [Kasper et al. (2000)].

The syntax checking code ConditonCheckingCode and ContextCheckingCode is used, e.g., by vitCheckFormat/1 (see vitCheck.pl). The callable predicates are either Prolog built-ins, library predicates, defined in vitValues.pl or already explained in previous sections of this documentation.

The context list ContextOfConditon is given by the class description. Together with parts of the checking code like vitGroupLabel/1 this information is used in vitCheckContent/1. In general, this kind of information is optional information which may occur together with the condition in a VIT.

The list of sortal restrictions SortalRestrictions is not used otherwise within the ADT package but can be used during VIT construction.

Optional subgoals Goals are called when accessing a lexicon.

vitValues.pl contains predicates defining value ranges for designators as described in [Kasper et al. (2000)] or in the original databases.

Note, that every lexicon access is deterministic. I.e. once a matching entry was found no choice points remain and no other entries will be tried.

Format of a Linking Table

    The linking tables are indexed in two ways: by semantic predicate names and syntactic lemmata. The format is

Language:vitSemLinking(PredName,Lemma,Particles,Stem,Linking) :-
   Subgoals.

Language:vitSynLinking(Lemma,Particles,Stem,PredName,Linking) :-
   Subgoals.

Language defines a language specific module.

PredName is the name of the predicate given by the semantic databases.

Lemma is also given by the semantic databases.

Particles is a list of separable particles or prefixes belonging to a word stem Stem.

Linking is a list with elements of the form arg(SemRole, Case, GramFunc, ArgType) and with arguments as follows:

1.

a semantic role SemRole or maybe a list of roles if the clause corresponds to an ambiguous verb;

2.

a case Case or maybe a list of case values which corresponds to a list of grammatical functions, repectively. In case of a subcategorized preposition (SemRole=arg4), the value is always noc, for ``no case'';

3.

a grammatical function GramFunc or maybe a list of function values which corresponds to a list of case values, repectively. In case of a subcategorized preposition, the preposition value or a list of possible ones. If no specific value can be determined, the default is nof for ``no function'';

4.

an argument type ArgType. The default is np if in the original database does not specify a value.

For further value specifications see Section  or the original Prolog file.

Access to Linking Tables

The ADT package provides the following access predicates to the linking tables:

vitGramRole(+Language,+Predicate,+GRole,-SArg) 

Given a language, the name of a semantic predicate and a grammatical role, vitGramRole/4 returns the corresponding semantic argument role.

Example:

| ?- vitGramRole(ge,absagen,subj,Arg).

Arg=arg1

| ?- vitGramRole(ge,absagen,obj2,Arg).

Arg=arg2

Adding Further Lexicon Entries

The ADT packages comes with hook predicates which makes it very easy to add new entries to the on-line version. The supported languages are English, German, and Japanese whereas the lexicons can be found in modules en, de, and jp, respectively. All additional lexicon entries should have the format described in Section .

First, each of these modules checks for lexical entries in a module named language _first, where language is one of en, de, and jp depending on the selected language. Then the original lexicon of the ADT package is looked up. Finally, each of these language modules checks for lexical entries in a module named language _last. Hence, on the one hand, one can add entries at the beginning of a lexicon using module language _first to overide existing entries. On the other hand, adding at the end of the existing lexicons is used to enlarge the lexicon. The named hook predicates are all assumed to be declared multifile and dynamic.

Beside these hooks, all user defined lexicon modules are considered when a user switches manually to a language using vitSetMyLanguage/1 (see Section ).

Note that both, the lexicon hooks and the user defined language modules, only extend the possible contents of the Conditions slot. The remaining information in a VIT is assumed to be known.