1. Representing Specialized Events
with FrameBase
Jacobo Rouces
Aalborg University
jrg@es.aau.dk
Gerard de Melo
Tsinghua University
gdm@demelo.org
Katja Hose
Aalborg University
khose@cs.aau.dk
2. 06/01/15 Rouces, De Melo, Hose – FrameBase 2
Overview
● FrameBase
● Integration of events: Dbpedia
● Integration of events: schema.org
● Integration of event aspects
● Complexity
● Representional flexibity
● Conclusions
3. 06/01/15 Rouces, De Melo, Hose – FrameBase 3
FrameBase: motivation
● Different KBs use different ways to represent N-ary
relations
– Using Direct Binary Relations
● Used as “default” mode in most KBs. Dereified.
– RDF reification
● YAGO,YAGO2s
– Subproperties
● Proposed in [Nguyen et al, WWW 2014]
– Neo-davidsonian representations
● To an extent used in most Kbs that include events. E.g. Freebase
● Difficult to link (no equivalence relations) and to query.
4. 06/01/15 Rouces, De Melo, Hose – FrameBase 4
FrameBase: schema
● Core: RDFS schema to represent knowledge using neo-
Davidsonian approach with a wide and extensible vocabulary of
– frames (events, situations, frames, eventualities…)
– frame elements (outgoing properties representing frame-specific
semantic roles)
● Vocabulary based on NLP resources (FrameNet+WordNet)
– This provides connection with natural language and semantic role labeling
systems (e.g. detect events with SEMAFOR). Cluster near-equivalents.
● Inference rules to provide direct binary predicates
?f a :frame-Separating-partition.v
?f :fe-Separating-Whole ?s ?s :isPartitionedIntoParts ?o
?f :fe-Separating-Parts ?o
5. 06/01/15 Rouces, De Melo, Hose – FrameBase 5
FrameBase: schema
e1
Frame type
e2
e3
FRAME CLASS
FRAME ELEMENT
(FRAME-SPECIFIC
SEMANTIC ROLES)
FRAME INSTANCE
DIRECT BINARY
PREDICATE
6. 06/01/15 Rouces, De Melo, Hose – FrameBase 6
FrameBase: ReDer rules
● Two-layered structure:
☞Create two levels of reification, and reification-dereification
(ReDer) inference rules that connect them.
● Reified knowledge using frames and frame elements
● Dereified knowledge using direct binary predicates
– Rules are Horn clauses (good for inference engines)
e1
Event type
e2
e3
?f a :frame-Separating-partition.v
AND
?f :fe-Separating-Whole ?s
AND
?f :fe-Separating-Parts ?o
IFF
?s ..-isPartitionedIntoParts ?o
7. 06/01/15 Rouces, De Melo, Hose – FrameBase 7
FrameBase: example
yago:Nobel_Prize
...-competitor
yago:Nobel_Prizeyago:Nobel_Prize
fe-Win_prize-competition
fe-Win_prize-prize
1921^xsd:date
...-time ...-explanation
BEYOND TIME
AND LOCATION!
8. 06/01/15 Rouces, De Melo, Hose – FrameBase 8
FrameBase: example
:frame-Win_prize-win.v
...-competitor
yago:A_Einsteinyago:Nobel_Prize
fe-Win_prize-competition
fe-Win_prize-prize
1921^xsd:date
...-time
yago:Photoelectric_effect
...-explanation
frame:Working_on-work.n
fe-Working_on-agent
...-domain
...-time
1905^xsd:date
BEYOND TIME
AND LOCATION!
9. 06/01/15 Rouces, De Melo, Hose – FrameBase 9
FrameBase: example
:frame-Win_prize-win.v
...-competitor
yago:A_Einsteinyago:Nobel_Prize
fe-Win_prize-competition
fe-Win_prize-prize
1921^xsd:date
...-time
?
?
?
yago:Photoelectric_effect
...-explanation
frame:Working_on-work.n
fe-Working_on-agent
...-domain
...-time
1905^xsd:date
?
BEYOND TIME
AND LOCATION!
10. 06/01/15 Rouces, De Melo, Hose – FrameBase 10
FrameBase: example
:frame-Win_prize-win.v
...-competitor
yago:A_Einsteinyago:Nobel_Prize
fe-Win_prize-competition
fe-Win_prize-prize
1921^xsd:date
...-time
winsByCompetitor
winsAtTime
isWonAtTime
yago:Photoelectric_effect
...-explanation
frame:Working_on-work.n
fe-Working_on-agent
...-domain
...-time
1905^xsd:date
worksAtTime
BEYOND TIME
AND LOCATION!
11. 06/01/15 Rouces, De Melo, Hose – FrameBase 11
FrameBase: ReDer rules
● FrameBase: Two-layered structure:
☞Create two levels of reification, and inference rules that
connect them.
● Reified knowledge using frames and frame elements
● Dereified knowledge using direct binary predicates
– Rules are Horn clauses (good for inference engines)
– Around 15000 rules and
direct binary predicates are
created automatically.
– Different storage strategies
are possible.
?f a :frame-Separating-partition.v
AND
?f :fe-Separating-Whole ?s
AND
?f :fe-Separating-Parts ?o
IFF
?s ..-isPartitionedIntoParts ?o
12. 06/01/15 Rouces, De Melo, Hose – FrameBase 12
FrameBase: integration rules
● Integration rules from source KBs can be created with
SPARQL CONSTRUCT queries (and optionally a RDFier)
CONSTRUCT {
_:e a framebase:frame-People_by_jurisdiction-citizen.n .
_:e framebase:fe-People_by_jurisdiction-Person ?person .
_:e framebase:fe-People_by_jurisdiction-Jurisdiction ?country .
} WHERE {
?person freebase:people.person.nationality ?country .
}
● More examples in the ESWC 2015 paper “FrameBase:
Representing N-ary Relations Using Semantic Frames”
14. 06/01/15 Rouces, De Melo, Hose – FrameBase 14
Integration of events: DBpedia
#For sub-classes of dbpedia-owl:Event
CONSTRUCT {
?e a :frame-Social_event-meeting.n .
} WHERE {?e a dbpedia-owl:SocietalEvent}
#For sub-classes of dbpedia-owl:SocietalEvent
CONSTRUCT {
?e a :frame-Project-project.n .
?e :fe-Project-Activity dbpedia:Space_exploration .
} WHERE {?e a dbpedia-owl:SpaceMission}
#For sub-classes of dbpedia-owl:SocietalEvent
CONSTRUCT {
?e a fbe:frame-Social_event-convention.n .
} WHERE {?e a dbpedia-owl:Convention}
SHOULD MATCH THE SUPERCLASS RULE TOO
17. 06/01/15 Rouces, De Melo, Hose – FrameBase 17
Integration of event aspects
● Time and space:
– Frame elements ...-Time and ...-Place
:fe-Social_event-Place
:fe-Social_event-Time
:fe-Competition-Place
:fe-Competition-Time
:fe-Smuggling-Place
:fe-Smuggling-Time
...
18. 06/01/15 Rouces, De Melo, Hose – FrameBase 18
Integration of event aspects
● Participation
– Frames include participants like agents, patients, etc.
:fe-Commerce_buy-Buyer → :fe-Getting-Recipient
:fe-Destroying-Destroyer → :fe-Transitive_action-Agent
:fe-Destroying-Undergoer → :fe-Transitive_action-Patient
19. 06/01/15 Rouces, De Melo, Hose – FrameBase 19
Integration of event aspects
● Relations between events
– Mereology (part of)
● Using a suitable FE, when available (see example):
?whole :fe-Social_event-Occasion ?part
● Or using one frame and 2 FEs:
_:i a :frame-Part_whole .
_:i :fe-Part_whole-Part ?part .
_:i :fe-Part_whole-Whole ?whole .
● Both options can be used
20. 06/01/15 Rouces, De Melo, Hose – FrameBase 20
Integration of event aspects
● Relations between events
– Causality (cause of)
● Using a suitable FE, when available (see example):
?consequence :fe-Event-Reason ?cause
● Or using one frame and two FEs:
_:i a :frame-Causation .
_:i :fe-Causation-Cause ?cause .
_:i :fe-Causation-Effect ?consequence .
● Both options can be used
– Correlation (share common cause)
● Instantiate a common cause using the methods above
21. 06/01/15 Rouces, De Melo, Hose – FrameBase 21
Integration of event aspects
● Documentation
– Events can be “documented using some media like photos or videos captured during the event”
(Scherp and Mezaris 2014)
– Using the frame Recording with an available lexical unit (like recording.v) or an extension:
_:i a :frame-Recording-record.v .
_:i :fe-Recording-Phenomenon ?event .
_:i :fe-Recording-Medium ?media .
Medium is the physical entity in which the Agent creates a record of their
impression of the Phenomenon.
In fact, Pepys recorded everything in his diary.
22. 06/01/15 Rouces, De Melo, Hose – FrameBase 22
Integration of event aspects
● Interpretation
– Very broad definition: “capturing subjectivity that may exist on the other aspects of events”
(Scherp and Mezaris 2014)
● Using perspectivization from FrameNet
:frame-Commerce_sell → :frame-Commerce
:frame-Commerce_buy → :frame-Commerce
● Using extra frame, e.g.:
_:i a :frame-Becoming_Aware .
_:i :fe-Becoming_Aware-Cognizer ...
_:i :fe-Becoming_Aware-Instrument …
_:i :fe-Becoming_Aware-Means ...
_:i :fe-Becoming_Aware-Phenomenon ...
The Cognizer is the person who becomes aware of a Phenomenon
Pat discovered a great little restaurant in Soho.
The Cognizer uses an Instrument to (enable themselves to) become
aware of the Phenomenon.
Olson says the deputy came out to the farm and detected a radioactive
substance with his Geiger counter.
23. 06/01/15 Rouces, De Melo, Hose – FrameBase 23
Complexity
● Basic structure of integration rule
SOURCE FRAMEBASE
Event Frame class
Outgoing property 1 Frame element 1
Outgoing property 2 Frame element 2
... ...
Outgoing property n Frame element n
24. 06/01/15 Rouces, De Melo, Hose – FrameBase 24
Complexity
● Deviations: several frames instantiated
SOURCE FRAMEBASE
Event Frame class 1 ... Frame class n
Outgoing property 1 Frame element 1,1 ... Frame element 1,n
Outgoing property 2 Frame element 2,1 .. Frame element 2,n
... ... ... ...
Outgoing property n Frame element n,1 .. Frame element n,n
25. 06/01/15 Rouces, De Melo, Hose – FrameBase 25
Complexity
● Deviations: frame elements inverted
SOURCE FRAMEBASE
Event Frame class
Outgoing property 1 Frame element 1 ^Frame element 1
Outgoing property 2 Frame element 2 ^Frame element 2
... ... ...
Outgoing property n Frame element n ^Frame element n
26. 06/01/15 Rouces, De Melo, Hose – FrameBase 26
Complexity
● Deviations: property in source (reification)
SOURCE FRAMEBASE
Property Frame class
Frame element 1
Frame element 2
...
Frame element n
27. 06/01/15 Rouces, De Melo, Hose – FrameBase 27
Complexity
● All these deviations, and possibly others,
can be combined
28. 06/01/15 Rouces, De Melo, Hose – FrameBase 28
Representational flexibility
● Two ways of “narrowing down”
– New lexical unit
X a :frame-Statement-twit.n
– Assign a value to a frame element
X a :frame-Statement-write.v
X :fe-Text_creation-Place dbpedia:Twitter
29. 06/01/15 Rouces, De Melo, Hose – FrameBase 29
Representational flexibility
● Further reification
– Remember different options for mereology, causality,
interpretation...
– Even for participation: There is a frame Participation
– It boils down to the fact that FEs can be reified
(substituted by a frame instance and two FEs), and
sometimes FrameNet/FrameBase have only one option
and sometimes have both.
– Possible solution: further reification-dereification rules?
30. 06/01/15 Rouces, De Melo, Hose – FrameBase 30
Conclusions
● FrameBase offers wide-range, natural-language-related
and extensible schema for representation of events and
more...
● But automatic creation of rules remains a challenge:
– But rules are complex, which makes any training space
very sparse, which makes difficult to train any system to
create them automatically.
– Still some heterogeneity
31. 06/01/15 Rouces, De Melo, Hose – FrameBase 31
Data
● More information: http://framebase.org
● Data is open-source.
– License: CC-BY 4.0 International
The research leading to these results has received funding from the European Union
Seventh Framework Programme (FP7/2007-2013) under grant agreement No. FP7-SEC-
2012-312651 (ePOOLICE project).
Additional funding was provided by National Basic Research Program of China Grants
2011CBA00300, 2011CBA00301, and NSFC Grants 61033001, 61361136003.