Machine-annotated Rationales: Faithfully Explaining Text Classification

Elize Herrewijnen; Dong Nguyen; Jelte Mense; Floris Bex

Machine-annotated Rationales: Faithfully Explaining Text Classification

Research output: Contribution to conference › Paper › Academic

Abstract

We propose an approach to faithfully explaining text classification models, using a specifically designed neural network to
find explanations in the form of machine-annotated rationales
during the prediction process. This results in faithful explanations that are similar to human-annotated rationales, while not
requiring human explanation examples during training. The
quality of found explanations is measured on faithfulness,
quantitative similarity to human explanations, and through a
user evaluation.

Original language	English
Number of pages	8
Publication status	Published - 2021
Event	35th AAAI Conference on Artificial Intelligence - Duration: 8 Feb 2021 → 9 Feb 2021

Conference

Conference	35th AAAI Conference on Artificial Intelligence
Period	8/02/21 → 9/02/21

Cite this

@conference{fae9038a3d6240a2b38d87e8bdf7e8f8,

title = "Machine-annotated Rationales: Faithfully Explaining Text Classification",

abstract = "We propose an approach to faithfully explaining text classification models, using a specifically designed neural network to find explanations in the form of machine-annotated rationales during the prediction process. This results in faithful explanations that are similar to human-annotated rationales, while not requiring human explanation examples during training. The quality of found explanations is measured on faithfulness, quantitative similarity to human explanations, and through a user evaluation.",

author = "Elize Herrewijnen and Dong Nguyen and Jelte Mense and Floris Bex",

year = "2021",

language = "English",

note = "35th AAAI Conference on Artificial Intelligence ; Conference date: 08-02-2021 Through 09-02-2021",

}

TY - CONF

T1 - Machine-annotated Rationales: Faithfully Explaining Text Classification

AU - Herrewijnen, Elize

AU - Nguyen, Dong

AU - Mense, Jelte

AU - Bex, Floris

PY - 2021

Y1 - 2021

N2 - We propose an approach to faithfully explaining text classification models, using a specifically designed neural network to find explanations in the form of machine-annotated rationales during the prediction process. This results in faithful explanations that are similar to human-annotated rationales, while not requiring human explanation examples during training. The quality of found explanations is measured on faithfulness, quantitative similarity to human explanations, and through a user evaluation.

AB - We propose an approach to faithfully explaining text classification models, using a specifically designed neural network to find explanations in the form of machine-annotated rationales during the prediction process. This results in faithful explanations that are similar to human-annotated rationales, while not requiring human explanation examples during training. The quality of found explanations is measured on faithfulness, quantitative similarity to human explanations, and through a user evaluation.

M3 - Paper

T2 - 35th AAAI Conference on Artificial Intelligence

Y2 - 8 February 2021 through 9 February 2021

ER -

Machine-annotated Rationales: Faithfully Explaining Text Classification

Abstract

Conference

Fingerprint

Cite this