Difference between revisions of "WordNet - RTE Users"

From ACL Wiki
Jump to navigation Jump to search
Line 35: Line 35:
 
|  
 
|  
 
| FIRST USE: Argument alignment between T and H.<br/>SECOND USE: used to change all the nominal predicates into verbs, to calculate relatedness between T and H (using VerbOcean).
 
| FIRST USE: Argument alignment between T and H.<br/>SECOND USE: used to change all the nominal predicates into verbs, to calculate relatedness between T and H (using VerbOcean).
| FIRST USE: Ablation test performed. Impact of the resource: -0.17% accuracy/null respectively on two-way and three-way task for run1; +0.16%/+0.34% for run2; +0.17%/+0.17% for run3.<br/>SECOND USE (WordNet+VerbOcean): null/+0.17% accuracy respectively on two-way and three-way task for run1; +0.5%/+0.67% for run2; +0.17%/+0.17% for run3.
+
| FIRST USE: Ablation test performed. Impact of the resource: -0.17% accuracy/null respectively on two-way and three-way task for run1; +0.16%/+0.34% for run2; +0.17%/+0.17% for run3.<br/>
 +
SECOND USE (WordNet+VerbOcean): null/+0.17% accuracy respectively on two-way and three-way task for run1; +0.5%/+0.67% for run2; +0.17%/+0.17% for run3.
  
 
|- bgcolor="#ECECEC" align="left"
 
|- bgcolor="#ECECEC" align="left"
Line 48: Line 49:
 
| RTE5
 
| RTE5
 
|  
 
|  
| Similarity between lemmata, computed by WordNet-based metrics  
+
| FIRST USE: Similarity between lemmata, computed by WordNet-based metrics.<br/>
| Ablation test performed. Positive impact of the resource on two-way run: +0.83% accuracy. Negative impact  on three-way run: -0.33% accuracy (-0.5% for two-way derived).
+
SECOND USE: antonymy relations between verbs.
 +
| FIRST USE: Ablation test performed. Positive impact of the resource on two-way run: +0.83% accuracy. Negative impact  on three-way run: -0.33% accuracy (-0.5% for two-way derived).<br/>
 +
SECOND USE (WordNet+VerbOcean+DLSIUAES_negation_list): positive impact on two-way run: +0.66% accuracy. Negative impact  on three-way run: -1% (-0.5% for two-way derived).
  
 
|- bgcolor="#ECECEC" align="left"
 
|- bgcolor="#ECECEC" align="left"

Revision as of 10:04, 1 December 2009

When not otherwise specified, the data about version, usage and evaluation of the resource have been provided by participants themselves.

Participants* Campaign Version Specific usage description Evaluations / Comments
AUEB RTE5 During the calculation of the similarity measures we treat words from T and H that are synonyms according to WordNet as identical. Ablation test performed. Negative impact of the resource: -2% accuracy on two-way, -2.67% on three-way task.
BIU RTE5 3.0 Synonyms, hyponyms (2 levels away from the original term), the hyponym_instance relation and derivations. Ablation test performed. Positive impact of the resource: +2.5% accuracy on two-way task.
Boeing RTE5 The system makes uses Wordnet synonyms, hypernyms relationships between (senses of) words, "similar" (SIM), "pertains" (PER), and "derivational" (DER) links to recognize equivalence between T and H. Ablation test performed. Positive impact of the resource: +4% accuracy on two-way, +5.67% on three-way task.
DFKI RTE5 FIRST USE: Argument alignment between T and H.
SECOND USE: used to change all the nominal predicates into verbs, to calculate relatedness between T and H (using VerbOcean).
FIRST USE: Ablation test performed. Impact of the resource: -0.17% accuracy/null respectively on two-way and three-way task for run1; +0.16%/+0.34% for run2; +0.17%/+0.17% for run3.

SECOND USE (WordNet+VerbOcean): null/+0.17% accuracy respectively on two-way and three-way task for run1; +0.5%/+0.67% for run2; +0.17%/+0.17% for run3.

DirRelCond RTE5 Use of many WordNet relations (such as synonymy, hypernymy, hyponymy, meronymy, holonymy etc.) to compute the relatedness between words with the same part of speech in T and H. No ablation test performed. The resource cannot be removed without breaking the functionality of the system.
DLSIUAES RTE5 FIRST USE: Similarity between lemmata, computed by WordNet-based metrics.

SECOND USE: antonymy relations between verbs.

FIRST USE: Ablation test performed. Positive impact of the resource on two-way run: +0.83% accuracy. Negative impact on three-way run: -0.33% accuracy (-0.5% for two-way derived).

SECOND USE (WordNet+VerbOcean+DLSIUAES_negation_list): positive impact on two-way run: +0.66% accuracy. Negative impact on three-way run: -1% (-0.5% for two-way derived).

AUEB RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
BIU RTE4 3.0 Synonyms, hyponyms (2 levels away from the original term), the hyponym_instance relation and derivations. Also used as part of our novel lexical-syntactic resource 0.8% improvement in ablation test on RTE-4. Potential contribution is higher since this resource partially overlaps with the novel lexical-syntactic rule base
Boeing RTE4 2.0 Semantic relation between words No formal evaluation. Plays a role in most entailments found
Cambridge RTE4 3.0 Meaning postulates from WordNet noun hyponymy, e.g. forall x: cat(x) -> animal(x) No systematic evaluation
CERES RTE4 3.0 Hypernyms, antonyms, indexWords (N,V,Adj,Adv) Used, but no evaluation performed
DFKI RTE4 3.0 Semantic relation between words No separate evaluation
DLSIUAES RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
EMORY RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
FbkIrst RTE4 3.0 Lexical similarity No precise evaluation of the resource has been carried out. In our second run we used a combined system (EDITSneg + EDITSallbutneg), and we had an improvement of 0.6% in accuracy with respect to the first run in which only EDITSneg was used. EDITSallbutneg exploits lexical similarity (WordNet similarity), but we can’t affirm with precision that the improvement is due only to the use of WordNet
FSC RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
IIT RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
IPD RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
OAQA RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
QUANTA RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
SAGAN RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
Stanford RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
UAIC RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
UMD RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
UNED RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
Uoeltg RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
UPC RTE4 Data taken from the RTE4 proceedings. Participants are recommended to add further information.
AUEB RTE3 2.1 Synonymy resolution Replacing the words of H with their synonyms in T: on RTE3 data sets 2% improvement
UIUC RTE3 Semantic distance between words
VENSES RTE3 3.0 Semantic relation between words No evaluation of the resource
New user Participants are encouraged to contribute.
Total: 24


[*] For further information about participants, click here: RTE Challenges - Data about participants

   Return to RTE Knowledge Resources