e-SNLI: Natural language inference with natural language explanations
In order for machine learning to garner widespread public adoption, models must be able to provide interpretable and robust explanations for their decisions, as well as learn from human-provided explanations at train time. In this work, we extend the Stanford Natural Language Inference dataset with...
Main Authors: | , , , |
---|---|
Format: | Conference item |
Published: |
Neural Information Processing Systems
2018
|
_version_ | 1797060225727987712 |
---|---|
author | Camburu, O Rocktäschel, T Lukasiewicz, T Blunsom, P |
author_facet | Camburu, O Rocktäschel, T Lukasiewicz, T Blunsom, P |
author_sort | Camburu, O |
collection | OXFORD |
description | In order for machine learning to garner widespread public adoption, models must be able to provide interpretable and robust explanations for their decisions, as well as learn from human-provided explanations at train time. In this work, we extend the Stanford Natural Language Inference dataset with an additional layer of human-annotated natural language explanations of the entailment relations. We further implement models that incorporate these explanations into their training process and output them at test time. We show how our corpus of explanations, which we call e-SNLI, can be used for various goals, such as obtaining full sentence justifications of a model’s decisions, improving universal sentence representations and transferring to out-of-domain NLI datasets. Our dataset1 thus opens up a range of research directions for using natural language explanations, both for improving models and for asserting their trust. |
first_indexed | 2024-03-06T20:14:20Z |
format | Conference item |
id | oxford-uuid:2ba47384-691f-4fab-b5a3-9770278888d3 |
institution | University of Oxford |
last_indexed | 2024-03-06T20:14:20Z |
publishDate | 2018 |
publisher | Neural Information Processing Systems |
record_format | dspace |
spelling | oxford-uuid:2ba47384-691f-4fab-b5a3-9770278888d32022-03-26T12:32:06Ze-SNLI: Natural language inference with natural language explanationsConference itemhttp://purl.org/coar/resource_type/c_5794uuid:2ba47384-691f-4fab-b5a3-9770278888d3Symplectic Elements at OxfordNeural Information Processing Systems2018Camburu, ORocktäschel, TLukasiewicz, TBlunsom, PIn order for machine learning to garner widespread public adoption, models must be able to provide interpretable and robust explanations for their decisions, as well as learn from human-provided explanations at train time. In this work, we extend the Stanford Natural Language Inference dataset with an additional layer of human-annotated natural language explanations of the entailment relations. We further implement models that incorporate these explanations into their training process and output them at test time. We show how our corpus of explanations, which we call e-SNLI, can be used for various goals, such as obtaining full sentence justifications of a model’s decisions, improving universal sentence representations and transferring to out-of-domain NLI datasets. Our dataset1 thus opens up a range of research directions for using natural language explanations, both for improving models and for asserting their trust. |
spellingShingle | Camburu, O Rocktäschel, T Lukasiewicz, T Blunsom, P e-SNLI: Natural language inference with natural language explanations |
title | e-SNLI: Natural language inference with natural language explanations |
title_full | e-SNLI: Natural language inference with natural language explanations |
title_fullStr | e-SNLI: Natural language inference with natural language explanations |
title_full_unstemmed | e-SNLI: Natural language inference with natural language explanations |
title_short | e-SNLI: Natural language inference with natural language explanations |
title_sort | e snli natural language inference with natural language explanations |
work_keys_str_mv | AT camburuo esnlinaturallanguageinferencewithnaturallanguageexplanations AT rocktaschelt esnlinaturallanguageinferencewithnaturallanguageexplanations AT lukasiewiczt esnlinaturallanguageinferencewithnaturallanguageexplanations AT blunsomp esnlinaturallanguageinferencewithnaturallanguageexplanations |