Discussion of open issues

Cases of meaning equivalence• Creative Language use

– Metaphorical language• Al7ayah ba2a lwnha bamby 5ales [Egyptian Arabic BOLT data]• [gloss] the-life became colored pink extremely• [Trans1] Life is so pink• [Trans2] Everything is extremely hunky dory• [Trans3] Life is quite rosy

– Sarcasm• Great, now I have to be at work at 5am• Great, now I have to be at work at 10am

– Etc.– Pronominal Usage

• He ate the apple• She ate the apple

– The notion of salience and the role it plays with respect to context (the IBM example, Sameer’s presentation)

Current Tools and Resources

• Current semantic technologies– WSD, WSI, Lexical substitution, SRL, modality,

distributional compositionality, MWE, scoping, conversion from surface form to logical structure

• Current resources– WN, ontologies (SUMO), MRD, ontonotes, UVI,

verbnet, propbank, framenet

What other things are needed?

• Tools– Creative language use detectors (work by Veale,

Muresan, also some by us here on sarcasm)• Metaphor, irony, sarcasm

– Register/style/salience detectors• Resources– Generalizable predicate relations detectors (extend

the UVI to be more comprehensive)– Annotations for pragmatic stuff – What data do we annotate? (Multi)MASC

What platform to adopt?

• Should we go with something like UIMA, or with Nancy’s grant– We can probably start with UIMA and then

transfer over especially if we want webservices and cloud computing capabilities

What semantic interoperability standards do we adopt?

• CAS• OWL• KAF• Etc.

Precursor issues

• Refine STS definition – Create an inventory of possible relations we observe between

two textual snippets– Devise set of diagnostics for each of these relations in the

inventory • Equivalence (substitutability)• Entailment• Contradiction• Specificity• Subset• etc

– What role does the Topic/domain play– What role does the context play– What role does the native language/culture play

Refine the STS Annotation Framework

• Adopt and work on Diana, Alessandro Lenci, Ido, Bernardo, Alon’s insights on how to modularize and concretize the task of STS for humans

• Three types of annotations– Cast as an alignment + scores– Decouple alignment from overall scores– Just scores

• How do we obtain the annotations– Turkers?– Trained people?

Evaluation Issues

• Intrinisic vs Extrinsic measures– What is desirable– How does it correlate with application needs

Should we attempt multilingual STS from the get go?

Next steps• Balancing act of functionality/utility/ontological issues• Create this community

– Committees to work on the different aspects • Shared Task

– *SEM as a venue– Use Nancy’s challenge idea

• Seek funding from different resources– NSF– DARPA– AirForce– IARPA

• Potential for European/Asian collaboration