Skip to main content
. Author manuscript; available in PMC: 2019 Dec 4.
Published in final edited form as: Proc Conf Empir Methods Nat Lang Process. 2019 Nov;2019:4240–4250. doi: 10.18653/v1/D19-1434

Table 2:

The easiest and hardest items judged by machine responses for each class in the SNLI test data set.

Premise Hypothesis Label Difficulty

Two men and a woman are inspecting the front tire of a bicycle. There are a group of people near a bike. Entailment −3.7
A girl in a newspaper hat with a bow is unwrapping an item. The girl is going to find out what is under the wrapping paper. Entailment 3.1

Two dogs playing in snow. A cat sleeps on floor Contradiction −4.0
Man sweeping trash outside a large statue. A man is on vacation. Contradiction 3.8

People sitting in chairs with a row flags hanging over them. A family reunion for Fourth of July Neutral −3.6
A group of dancers are performing. The audience is silent. Neutral 3.8