1. (2) Dipartimento di Scienze dell’Informazione, Università di Bologna(1) Semantic Technology Laboratory ISTC-CNR
Gathering Lexical Linked Data and
Knowledge Patterns from FrameNet
Andrea Giovanni Nuzzolese (1,2)
andrea.nuzzolese@istc.cnr.it
Aldo Gangemi (1)
aldo.gangemi@cnr.it
Valentina Presutti (1)
valentina.presutti@cnr.it
K-CAP 2011
Banff, AL, Canada
27 June 2011

2. Outline
• Motivations
• Semantic issues
• Transformation method
• Ongoing work
• Conclusions

3. Premise
• Work after request from Berkeley FrameNet group
for a Semantic Web version of FrameNet 1.5
• Previous work had various limitations, mainly data
incompleteness and implicit semantics
– E.g. Scheczyk et al., Narayanan et al.
• Decided to go for a dual transformation
– RDF for a complete porting to Linked Open Data,
similarly to W3C WordNet RDF porting
– (customizable) OWL for a focused porting to
knowledge patterns reusable for ontology design or
for creating views over linked data

4. Motivations
• The web of data is exploding and NLP techniques accompany
this explosion
• Hybridizing natural language processing and semantic web
techniques shows to be a promising approach
• Part of the exploitation of LOD data, is carried out by means
of lexical resources that are represented directly as linked
data
• Bring lexical resource on linked data (favor hybridization)
– benefit from linking all lexical resources and have an
homogenous more powerful one
• Link lexical knowledge to domain knowledge
– linked data ground to lexical knowledge and textual documents

5. DBpedia
Lexvo
lingvoj
RDF
WordNet
3.0
RDF
FrameNet
1.5
RDF
VerbNet
3.1
RDF
Italian
MultiWordNet
WordNet
Domains
WordNet
Supersenses
WordNet
Formal Glosses
VerbOcean

6. Several semantic issues
in reusable linguistic data
• Semantics induced by the data structure, e.g.
RDB, XML, etc.
• Semantics from the linguistic model adopted
• Semantics of the corpus (e.g. sentences)
• Semantics needed for querying
• Semantics needed for reasoning

7. FrameNet
• A lexical knowledge base
– cognitive soundness
– grounded in a large corpus
• Consists of a set of frames, which have
– frame elements
– lexical units, which pair words (lexemes) to frames
– relations to corpus elements
• Each frame can be interpreted as a class of
situations

8. An example of frame

9. FrameNet as LOD

10. FrameNet as LOD

11. FrameNet as ontologies

12. Structural
Schema
Linguistic
Schema
Linguistic
Data
Corpus
Data
Referential
Data
Linguistic
transformation
architecture

13. Transformation approach
• We pulled out the semantics of FrameNet and its
data by using Semion,
• Semion is a tool grounded on a method with two
main steps
– a syntactic and completely automatic transformation of the data source
to RDF datasets according to an OWL ontology that represents the data
source structure
– a semantic rule-based refactoring that allows to express the RDF triples
according to specific domain ontologies e.g. SKOS, DOLCE, FOAF,
LMM, or anything indicated by the user.

14. Reengineering
Syntactic transformation to RDF triples
<frame name="Abounding_with" ... ID="262">
...
<frameRelation type="Inherits from">
<relatedFrame>
Locative_relation
</relatedFrame>
</frameRelation>
...
</frame>

15. Refactoring
• aims to add semantics to data
• is performed by means of set of rules
– i.e. SPARQL CONSTRUCT

16. ABox Refactoring
The ABox refactoring is the
process of gathering RDF
data (Abox)
Rule-based
Customizable or based on
recipes

17. ABox Refactoring (data)

18. TBox Refactoring
• The TBox refactoring is the process of
gathering a new ontology schema (a
TBox) from data (ABox)

19. TBox Refactoring

20. Ongoing work
• Linking
– WordNet, WN Domains, MultiWordNet, VerbNet,
FrameNet, VerbOcean (P. Pantel)
• Basic linking uses SKOS
– exactMatch, closeMatch
– links partly present in Colorado bank, partly in
WordNet mappings, part are newly created
• More reasoning requires some expressivity
– semiotics.owl knowledge pattern, D&S
– property chains

21. Conclusion
• issues related to the conversion of lexical
resources
– more specifically to semantic issued of FrameNet
conversion
• a method to solve those issues (supported by
a tool)
• a conversion of FrameNet to RDF published
as a dataset in the LOD
• a method to convert FrameNet data into
knowledge patterns

22. Thank you
Andrea Nuzzolese
-
STLab, ISTC-CNR
&
Dipartimento di Scienze dell’Informazione
University of Bologna
Italy

23. 23
Semantic issues: objects
• Semantic frames/verb classes as twofold creatures
– intensional polymorphic relations (aka descriptions) + situation types
– Desiring(?experiencer, ?theme, ?time, ?loc, ?...)
• Frame elements/VN arguments as complex creatures
– (semantic) roles + concepts
• Semantic types are a mixture
– concepts, grammatical types, etc.
• Lexical units/VN class members as hybrid creatures
– lexically-oriented semantic frames
– bridges between semantic frames and word senses
– FN lex units belong to diverse parts of speech
• Annotated sentences contain syntactical realizations of semantic
frames (“exemplifications”)
– syntactic frames in VN, valences in FN
23

24. 24
Semantic issues: relations
• Inheritance in FN and VN is classic, can hold for situation types safely
– needs to be treated jointly with semantic role representation
– subFe also classic
• Subframes in FN are conceptual compositions (“parts of descriptions”
in D&S), intensional in nature
– similarly for “excludes” and “requires” holding for FE
• Frame “usage” in FN is partial inheritance, hard to digest for situation
types
• Selectional restrictions in VN maybe too tough for situation types
• Selectional preferences absent in resources, but probability would be
an added value
• Core vs. peripheral vs. unexpressed are interesting but tough:
“characteristic”, hidden optionality, etc.
24

25. Why a KP?
– a multidimensional
context model able to
capture descriptive,
informational, situational,
social, and formal
characters of knowledge.