Specification of Requirements/Valence and Ontological Mapping

Summary on Requirements on Valence and Ontology Mapping (Synthesis by PC)

The lexicon-ontology model must support the specification of the syntactic valence or subcategorization behaviour of a lexical entry as well as how this syntactic argument structure maps to the semantic argument structure of the predicate or class in the ontology it refers to.

The lexicon-ontology model must support the specification of valences of arity greater than two and support the specification of how this syntactic argument structure maps to the semantic argument structure of a non-binary predicate. The lexicon-ontology model must allow to specify the grammatical function of the syntactic argument and should (optionally) support the specification of a thematic role as a function of the syntactic argument. It must allow to specify any particles and prepositional markers that specify the subcategorization behaviour of the lexical entry.

Further, the lexicon-ontology model will not prescribe which type of syntactic arguments and grammatical functions exist, nor which thematic roles exist. The lexicon-ontology model must be agnostic in this respect to be compatible with very different linguistic, syntactic and semantic theories.

The general recommendation for such mappings is that lexical entry describing an event (verbs, deverbal nouns etc.) are mapped to an object that reifies the actual event, following a Neo-Davidsonian approach to the representation of the semantics of verbs and other lexical entries. Such a neo-davisonian approach in which a constant is introduced to represent the event and the arguments are specified through relations (object properties) between the event and the entity in question are anyway necessary for lexical entry with non-binary valences as OWL2 DL can not directly represent non-binary predicates.

"Valence and Ontology Mapping" describes the description of the frames of a particular word and how to derive ontological predicates from them.

Valence

Example (lemon):

knows is a transitive verb corresponding to foaf:knows

knows has two arguments: subject , and directObject

entry:knows lemon:synBehavior [ 
  a lemon:Frame ;
  lexinfo:subject _:subj ;
  lexinfo:directObject _:obj 
].

The subject is also the subject (in the domain) of foaf:knows, and the direct object is the object (range).

entry:knows lemon:sense [
  a LexicalSense ;
  lemon:reference foaf:knows ;
  lemon:subjOfProp _:subj ;
  lemon:objOfProp _:obj 
].

Open Issues

• What kind of arguments are there?
• Non-simple arguments: e.g., how do we indicate the preposition used in prepositional adjuncts?
• Semantic role arguments: e.g., 'theme', 'donor', 'recipient' à la FrameNet, VerbNet
• Full phrase tree: Should we support lexicalized grammars as frames (e.g., partial phrase trees with substitution points, à la LTAG grammars)

Mapping

Sticking to mapping to Triple structures, there are at most the following properties

• Subject of Property: Indicating the first place in the triple
• Object of Property: Indicating the third place in the triple
• Property: Indicating the second place in the triple (is this necessary?)
• Is A: Indicating membership to a class (i.e., first place of a rdf:type triple)

Higher-order Mapping

Mapping frames that may use several properties, use cases:

Events

"John gave Peter a book"

_:event1 a :GivingEvent ;
  :doner :John ;
  :recipient :Peter ;
  :theme [ a :Book ] ;
  :time "2002-09-24+06:00" .

Multi-element Relationships

"EEA agreement was made between EFTA and the EC"

:EEA a :TradeAgreeement ;
  :party :EFTA ;
  :party :EC .

Reified links

• From WikiData

Berlin's population is 3,499,879.

w:Berlin s:Population Berlin:Statement1 .
Berlin:Statement1 rdf:type o:Statement .
Berlin:Statement1 v:Population "3499879"^^xsd:integer .
Berlin:Statement1 q:As_of "2011-11-30"^^xsd:date .
Berlin:Statement1 q:Method w:Extrapolation .
Berlin:Statement1 rdfs:label "3,499,879 (As of Nov 30, 2011, Method Extrapolation)"@en .

These can be represented by simple queries

* s:Population ?city .
?city v:Population * .

How do we represent such queries?