search

uhh-lt / poincare

Using sense representations to improve detection of compositional nominal compounds
0 stars 0 forks source link

Compute the results on Reddy-Nouns #16

Open alexanderpanchenko opened 6 years ago

alexanderpanchenko commented 6 years ago

Please compute the results for the best models you computed so far (e.g. KR-50 dim) also for the Reddy-Nouns dataset (which contains only noun pairs from the Reddy dataset). The point is that Fahramand and Reddy++ where we get good results contains only noun phrases. We should thus exclude verb phrases to test the hypothesis that we are generalizing well across datasets (in the reasonable constrains - namely for the noun phrases).

https://docs.google.com/spreadsheets/d/1_fBfpgFLCOD6c7e39v-TSSjR8ThSmAaHsXdJg3yTCqA/edit#gid=0

alexanderpanchenko commented 6 years ago

@ajana1989 please upload a special version of the reddy dataset for nouns here https://github.com/uhh-lt/melting_pot/tree/master/Dataset_processed and provide the link to @dimitriusseveruscensor so he can accomplish this ticket.

alexanderpanchenko commented 6 years ago

this is the full 180 dataset

https://github.com/uhh-lt/melting_pot/blob/master/Dataset_processed/Reddy%2B%2B_180.txt