Please note that the examples below are preliminary versions of GRaSP. Updated examples will follow soon. Also note that the examples below go beyond the properties provided by GAF (attribution values to accommodate information on stances was not included in the original framework).
GRaSP examples (preliminary)
The examples below illustrate how we model attribution (who is the source of a claim, what is their commitment to the claim (is it certain/probable/possible), do they confirm or deny? etc.)
The main idea is that we maintain simple statements without this information on the instance level. Each mention may assign different attribution values to this statement (different people may claim the same thing, some sources may say something is true, others that it is false, etc.). The proposal is described in Beloki et al. (2015).
Simplified Illustration of how we model alternative perspectives on a statement.
Less simplified illustration of how we model (the same) alternative perspectives on this statement
Illustration of how we model alternative perspectives on Vaccinations
First complete GAF example (23 January 2014)
Please note the following:
- Provenance modeling for NLP modules is currently under development. Information related to provenance is subjected to change.
- The representations are simplified
- This example represents information on the Earthquake and Tsunami in Indonesia that took place on Christmas 2004.
- Two different sources (Veteran Today on the bottom left) and Bloomberg (the top graph) have different ideas about the cause of the catastrophe.
- The interpretation of the Bloomberg text is taken from two independent efforts of manual linguistic annotation on the same text: the TAF annotations provided by Sara Tonelli and Rachele Sprugnoli (Fondazione Bruno Kessler) and Colorado annotations provided by Kevin Crooks (University of Colorado Boulder). They are presented in the middle and bottom right graphs, respectively
- TAF annotations include information on participants. The Colorado annotations provide more fine-grained analyses between events. The example thus illustrates how output of different applications may be combined in GAF.
- The gaf:denotedBy relation can link an instance to a term in the text as well as relations between instances to specific linguistic relations between terms in the text.
The examples below were created while GAF was in its initial stages. They are present on this site in order to document how GAF was developed. They also provide more simplistic examples of GAF representations that are suitable for getting a basic impression. It should be noted that they may contain errors and certain relations are no longer used in GAF representations. Please use more recent examples as a reference to get an accurate impression of GAF in its current state.
Example 1, May 21st 2013, updated June 4th 2013
- A 9.1 temblor in 2004 caused a tsunami that swept across the Indian Ocean
- A graph gaf:G8 including the relation gaf:causes frome naacl:INSTANCE_186 to naacl:INSTANCE_188
- A taf:causal_construction from naacl:INSTANCE_MENTION_118 to naacl:INSTANCE_MENTION_120
- A graph gaf:G9 including the taf:CLINK mentioned above
- A relation sem:derivedFrom from gaf:G8 to gaf:G9
- An indication of provenance from gaf:G9 to taf:annotation_2013_03_24_id18681
Example 2, May 21st 2013
Two TAF annotations representing different interpretations of the sentence (GAF equivalences coming soon):
Indonesia’s West Papua province was hit by a magnitude 6.1 Earthquake today
- Interpretation 1: hit is a copula
- Interpretation 2: West Papua being hit is caused by the earthquake