ISOTC37SC4TDG6 Language Resource Ontologies

1
ISO/TC37/SC4/TDG6Language Resource Ontologies

• 2008-09-27, Pisa
• HASIDA Koiti
• hasida.k_at_aist.go.jp
• CfSR, AIST, Japan

2
Ontologization

• reformulation in terms of ontology
• provide standard way to convert annotations to
  labeled directed graphs
• DCR, LAF, LMF, FS, MAF, SemAF, SynAF, MLIF, etc.
• Cf. LMF and MAF have UML-based schemas.
• not XML but RDF as base description and modeling
  tool
• standard semantic interpretation for RDF
• highlight semantics rather than syntax

3
Purposes of Ontologization

• interoperability
• among ISO/TC37 standards
• with ontologies from elsewhere
• with any data containing linguistic content
• RDF data are easier to integrate than XML data.
• e.g. external annotation of texts in SMIL data
  without including linguistic description in SMIL
  specification
• fuller formalization of IS specifications
• semantic extension of DCR

4
Semantic Extension of DCR

• sorts of DCs
• unary predicate ? class
• binary relation ? property
• symmetric binary relation, etc.
• types of the domain (1st arg.) and the range (2nd
  arg.) of binary relations (properties)

5
XML Mess

• Semantic interpretation of XML is not
  standardized but defined ad hoc.
• Many inconsistent standards on overlapping
  issues.
• Huge standards containing many different semantic
  interpretation manners.
• e.g., MPEG-7 gt 2000 pages

6
RDF

• Resource Description Framework
• labeled directed graph
• W3C recommendation http//www.w3.org/RDF/
• Schemas are provided by RDFS, OWL, etc.
• textual representation
• XML, N3, etc.

7
RDF Graph
http//meetings.example.com/m1/hp
mhomePage
http//meetings.example.com/calm1
mattending
Fred
mgivenName
http//www.example.org/peoplefred
mhasEmail
mailtofred_at_example.com
8
Conversion of XML to RDF

• AnyURI- and IDREF(S)-type attribute
• ? object property (link)
• other attribute ? datatype property
• embedded element
• ? object/datatype property

9
24610 Feature Structure

• typed feature structure as in HPSG, etc.
• ISO 24610-1 Feature Structure Representation
• ISO 24610-2 Feature System Declaration
• labeled directed graph
• AVM (attribute-value matrix)
• textual encoding by XML

10
FS Graph RDF Graph
determiner
POS
ORTH
la

SPECIFIER
AGR


NUMBER
singular
AGR
HEAD
noun
POS

ORTH
pomme
11
FS in AVM
SPECIFIER HEAD
POS determiner ORTH la AGR 1NUMBER singular
POS noun ORTH pomme AGR 1
12
Ontologies Subsume Feature Systems

• Features are partial functions, whereas RDF
  properties are relations in general (possibly
  partial functions).
• Usual feature systems have no taxonomy of
  features, whereas usual ontologies have
  taxonomies of properties (e.g., due to
  rdfssubPropertyOf).

13
Feature-System Declaration
ltfsDecl type"word" baseTypes"sign"gt
ltfsDescrgtThe fundamental type for individual
wordslt/fsDescrgt ltfDecl name"orth"gt
ltfDescrgtThe orthographic representation for this
wordlt/fDescrgt ltvRangegtltstring/gtlt/vRangegt
lt/fDeclgt lt/fsDeclgt
The fundamental type for individual words
sign
rdfscomment
rdfssubClassOf
The orthographic representation for this word
word
rdfscomment
owlFunctionalProperty
rdftype
rdfsdomain
orth
string
rdfsrange
14
Constraint (Conditional)
ltcondgt ltfsgt ltf name"inv"gt ltbinary
value"true"/gt lt/fgt lt/fsgt ltthen/gt ltfsgt
ltf name"aux"gt ltbinary value"true"/gt
lt/fgt ltf name"vform"gt ltsymbol
value"fin"/gt lt/fgt lt/fsgt lt/condgt
inv
true
X
cond
aux
true
X
fin
vform
SWRL representation inv(?X,true) -gt aux(?X,true)
vform(?X,fin)
15
FS Ontologization (Summary)

• RDF ? FS
• Use ontologies for feature-system declarations.
• SWRL to encode constraints
• Defaults are outside of ontology.

16
24612 Linguistic Annotation Framework
17
GrAF in RDF
TOKEN
rdfstype
DET
POS
The
rdfstype
BASE
THE
clock
NN
POS
BASE
CLOCK

rdfstype
NP
NUMBER
SING
possibly stand-off annotation
18
SemAF-DActs
Dialogue
sender
1..
1..1
Turn
0..
Agent
overhearer
addressee
1..
1..
Utterance
0..
func.dep.
1..
DialogueAct
19
TODOs (projects in TDG6?)

• include ontologies in documents
• FSD
• just check UML (as far as no property hierarchy
  is necessary)
• LMF, MAF
• finish ontologization (possibly in UML)
• SynAF
• ontologize from scratch, forgetting XML
• DCR, SemAF-Time, SemAF-DActs, MLIF, etc.

20
Issues

• Who should ontologize individual WIs?
• ontologize future WIs from the beginning
• TDG6 should exemplify how.
• whether and how to make ontologization mandatory?
• Where to include ontologies of ongoing WIs?
• depending on their stages (WD, CD, ...)
• How to keep ontologizing DCs?
• replace DC metamodel by ontology?
• modify ISOCat?